Type your own name into a search engine. Ask an AI assistant who you are and what you do. Listen to how a host introduces you on a podcast. Each of those answers is assembled from the same raw material: the terms, entities, claims and questions that are currently linked to your name across everything that has been published, indexed and cited. That set is your semantic core. It exists whether you designed it or not.

Names are not ranked. Associations are.

Most people assume visibility is about being mentioned more. It is not. Search engines and AI models do not rank a name on its own. They rank the associations attached to it. If your name is strongly linked to a topic, a term or a category, you are visible for that topic. If it is not linked, you are invisible for it, regardless of how often you are mentioned elsewhere.

This explains a pattern we see constantly in the Audit. A founder with hundreds of mentions is still described by AI assistants using a job title from six years ago, because the strongest sources about them are old. A researcher with a landmark paper is not associated with the field the paper created, because the association was never stated plainly anywhere the models read.

The four parts of a core

Entities

The organisations, products, places and people you should be linked to. Your company. The category. The institution where the work happened. The partner you built it with. Entities are the strongest signal because they connect you to things that already have meaning.

Terms

The vocabulary a buyer or a journalist uses when they look for what you do. Not the words you like, the words they use. A core usually contains five to fifteen terms, and it is deliberately short so that each one can actually be owned.

Claims

The statements you can support with evidence. Not slogans. A claim is something a fact-checker could confirm: a result, a method, a first, a specific expertise. Claims are what get quoted.

Questions

The questions you should be the answer to. When a buyer asks an assistant how to choose between two approaches in your field, does your name come up? The questions are the most practical part of the core, because they define exactly what coverage has to do.

Why it comes before any campaign

Coverage without a core scatters. An article here, a podcast there, each on whatever the editor found interesting that week. Each piece may be good; together they say nothing consistent. Search and AI assistants weight consistent signals across independent sources far more than any single piece. That is why consistency beats volume and why every engagement we run starts with the Map.

Once the core exists, every opportunity can be filtered against it. Does this outlet carry the terms? Does this interview let you make the claim? Does this podcast sit in the field? If not, the answer is no, however attractive the placement looks. Deciding what to stop saying is half the work.

How to find yours

Start with the gap. Write down what you want to be known for in ten words. Then run the searches and prompts a buyer would run and write down what actually comes back. The distance between the two lists is your semantic gap. It is the most useful single page you can produce about your public presence, and it is where our work begins.

When someone asks an AI assistant about you, the answer is not looked up in a file with your name on it. It is assembled. Part of it comes from what the model learned during training, part from sources it retrieves at the moment of the question, and part from the way the question itself is phrased. Knowing which part is which is the difference between a plan that works and a plan that is wishful thinking.

Three inputs to every answer

Learned associations

During training a model reads an enormous amount of text and learns which terms appear near which names. If your name appeared repeatedly next to a category, a company and a set of claims, the model carries that pattern. If the pattern is weak or contradictory, the answer will be vague or wrong. This layer changes slowly and only through published, widely-read text.

Retrieved sources

Most assistants now search the web or a set of trusted sources before answering questions about specific people and companies. They lean heavily on a small number of source types: encyclopaedia-style pages, major publications, professional profiles, official sites and directories. What those pages say is what the assistant says. This layer can change within weeks, which is why directories and profiles matter more than most people think.

The question

Ask about a person in the context of one field and the assistant pulls the associations relevant to that field. Ask in another context and it pulls different ones. This is why we test a structured set of prompts rather than a single one, and why our reports show a pattern across repeated runs rather than one answer.

What this means for a campaign

The practical conclusion is that a campaign should aim at the sources the assistants read, with the associations you want them to learn, repeated consistently. That is a different target from raw reach. A feature in a title the models cite, an accurate professional profile, an episode page with a clean transcript and a correct description: each of these is worth more than a dozen mentions in places the models never look.

It also means that the measurement has to be honest. We record how the leading assistants describe a client before the work starts and at agreed checkpoints, run each prompt several times and report the pattern. We show the change. We do not promise a specific answer, because no one can. The method is described in our note on fair measurement.

Common mistakes

Publishing a great deal of content on your own site and expecting assistants to pick it up. Your own site is one source among many and it is discounted precisely because you wrote it. Independent, cited sources carry the weight.

Chasing mentions in outlets that are large but outside your field. The assistant learns the association between you and that outlet's topic, which may be the wrong one. Fit matters more than size, as explained in choosing a publication by fit.

Ignoring errors. An outdated title in a directory will be repeated by the assistant for years unless someone asks for it to be corrected. Assistants are confident even when they are wrong; the fix is upstream.

Start from the baseline

The first step is always to see what the assistants currently say. The Audit records it, source by source, prompt by prompt. Everything else follows from that baseline.

A worked example

A founder of a clinical software company asks an assistant what her company does. The answer names a product she discontinued two years ago and describes her by a former job title, because the strongest sources are a conference bio from 2023 and a directory entry copied from it. Neither is malicious; both are stale. The fix is not a press release. It is an updated profile, a corrected directory entry and two pieces in the trade titles the assistant cites, each stating the current product and the category term. Three months later the same prompt returns the current product in most runs. The old description still appears occasionally, which is what the pattern report will show, and which is why nobody should promise a single answer.

The lesson generalises. Every wrong answer has a source. Find the source, change the source, and give the assistant better material to read. That sequence is the whole method, and it is slow, unglamorous and measurable.

Every client conversation we have starts with the same page. On the left, what the client wants to be known for, in their own words. On the right, what search engines and AI assistants actually say. The distance between the two columns is the semantic gap. It is the output of the Audit, the input to the Map, and the honest basis for any plan.

Why the gap is usually wider than people expect

People know what they do. They forget that the public record was assembled by accident: an old profile, a conference bio written in a hurry, an article that focused on one project, a directory entry copied from another directory. None of it is wrong exactly. It is just out of date, incomplete and inconsistent, and search and AI systems repeat it faithfully.

There is also a timing problem. The record lags. The most-cited sources about a person are often several years old because they have had time to be linked and copied. What you did last year has not caught up yet. So the gap is partly a measure of how far the record is behind you.

Three kinds of gap

The missing association

You want to be known for a term and the record does not connect you to it at all. This is the most common gap and the most fixable, because it is filled by publishing the association in the right places.

The wrong association

The record connects you to something you no longer do, or never did. An old role, a former company, a misattributed project. This gap needs correction as well as new coverage, because the wrong signal keeps competing with the right one.

The contested association

You want a term that someone else owns more strongly. The gap here is not about you; it is about the field. Closing it may mean choosing a narrower term, co-authoring with the person who owns it, or accepting that this one is not worth the cost. Mapping the field is how you decide.

Reading a gap statement

Not every gap should be closed. A good statement ranks the gaps by two things: how much the association matters to the buyer you want, and how expensive it is to close. A missing association in a field where you have evidence and no strong competitor is cheap and valuable. A contested term owned by a global brand is expensive and possibly not worth it. The plan follows the ranking.

This is also where the sales conversation changes. Instead of choosing publications from a menu, the client sees which gaps matter and asks what closes them. The publication, the podcast or the byline becomes the answer to a question rather than a product on a list.

Measuring the gap over time

The gap statement is re-run at agreed checkpoints. The same searches, the same prompts, the same method. The change is shown side by side. AI answers vary and no one controls them, so we report patterns across repeated runs rather than single results. What we can show is whether the record has moved, and by how much. That is a measurement, not a promise, and it is the only honest way to report this work.

A short example

A research group wants to be known for a specific measurement technique it pioneered. The record says it is known for the university it sits in and the disease it studied first. The technique appears nowhere in the top results and in none of the assistant answers. That is a missing association, and it is cheap to close: a definition piece in the field's main journal-adjacent title, a talk with a clean episode page, an updated institutional profile that names the technique in the first line. Six months later the term appears in most assistant answers about the group, with the same three sources cited. The gap statement from the Audit and the checkpoint report sit side by side, and the difference is visible without commentary.

Not every gap closes this cleanly. A contested term against a large commercial player may not move at all. The statement says so up front, and the plan spends the budget where it will show.

A semantic core defines what you should be associated with. The semantic field is the territory around it: the other people and companies linked to the same terms, the publications and podcasts that carry them, the communities that discuss them and the sources that search engines and AI assistants cite when they answer questions in your category. The core tells you where to aim. The field tells you what is in the way and where the shortcuts are.

Mapping the field

We map the field in the Map using the same searches and prompts as the Audit, but pointed at the terms rather than at the client. For each term in the core we record who is currently associated with it, which sources make that association and how strong it is. The result is a picture of the competition for meaning, which is different from the competition for customers.

The two often diverge. A company can have a small market share and own a term completely because its founder wrote the defining article five years ago. A market leader can be invisible for the category term because its coverage is all about funding and hiring. The field map shows these mismatches, and they are where the opportunities are.

Four things the field tells you

Which terms are open

Some terms have no strong owner. The associations are scattered across many weak sources. These are the cheapest to take and the plan usually starts there.

Which terms are contested

Some terms are owned strongly by someone with more history and more coverage. Competing head-on is expensive. The choice is to narrow the term, to co-author or partner with the owner, or to leave it. All three are legitimate; pretending the owner does not exist is not.

Where the field is published

Every term has a home: a trade title, a conference, a podcast, a newsletter, a directory. These are the outlets that carry the association and the ones the assistants cite. A placement there is worth more than a larger placement elsewhere. This is the logic of choosing by fit.

Who the co-authors and partners are

The people already associated with your terms are not only competitors. Some are potential collaborators. A joint piece with someone who owns an adjacent term transfers part of that association to you faster than any amount of solo coverage.

From field to plan

With the field mapped, the plan almost writes itself. Open terms first, through the outlets where the field lives. Contested terms narrowed or partnered. Wrong associations corrected. Everything filtered against the core so nothing scatters. The placements, appearances and LinkedIn work that follow are all chosen because the field map says they will compound.

One more point. The field includes the sources AI assistants rely on, and those are not always the biggest names. A specialist directory or an association's member page can carry more weight in an answer than a national newspaper. The map records this, and the plan uses it.

How the map is used in a quarter

The field map is not a research report to be filed. It is the working document for the campaign quarter. Each open term gets an owner on the client side, a target outlet from the map and a piece scheduled. Each contested term gets a decision recorded: narrow, partner or leave, with the reason. Each source the assistants cite is checked for accuracy and, where it is a directory or profile, updated. At the quarter's end the map is re-run and the changes are noted: which terms moved, which competitors published, which outlets changed. The plan for the next quarter comes from those notes rather than from a fresh brainstorm.

This is what makes the work compound. A campaign without a field map restarts every quarter. A campaign with one builds on what it learned last time, and the record shows it.

A semantic core that sits in a document changes nothing. It starts to work when it is turned into two practical tools: a content architecture that tells every writer, every host and every account manager what to say, and a filter that tells everyone what to decline. Most of the value of the Map is realised in this step.

The content architecture

The architecture has three levels. Pillar topics are the two to four themes that carry the core; every piece of coverage belongs to one of them. Under each pillar sits a message hierarchy: the primary claim, the supporting claims, the evidence, the terms to use. Under that sit the questions the pillar should answer, which become article angles, podcast talking points and LinkedIn posts.

The point of the structure is repetition with variety. The same claims and terms appear across many pieces in different forms, so search and AI assistants receive a consistent signal from independent sources. This is why consistency beats volume. Without the architecture, each piece is written from scratch and the signal scatters.

The filter

The filter is a short list of questions applied to every opportunity. Does the outlet carry the terms? Does the format let us make the claim? Does the audience include the buyer or the people who influence the buyer? Is this in the field or adjacent to it? Does it require us to say something outside the core? A no on any of these is usually a no on the opportunity.

What to stop saying

This is the part clients find hardest. Every company has phrases it has used for years that no longer serve the core. A founder who still describes the company by its first product. A researcher whose bio leads with an institution they left. A brand that mentions three unrelated capabilities in every paragraph because each one has a champion internally.

The exclusion list from the Map names these directly. It is not censorship; it is focus. Every term you keep using competes with the ones you want to own, and the record only has so much room. Removing a phrase from your bio, your site and your pitch is often the fastest single change you can make.

Applying it inside the company

The architecture and filter are handed to everyone who speaks or writes publicly: the founder, the marketing team, the agency, the person who updates the directory entries. Consistency across all of them is the goal. In-house teams use it to brief agencies; agencies use it under white-label arrangements to brief their own writers.

The test of whether it has been applied is simple. Take any ten pieces published in the last quarter and check them against the core. If eight or more carry the pillar, the claim and the terms, the filter is working. If not, the filter is being bypassed, usually by someone who found the opportunity too good to decline.

Revisiting the structure

The architecture is reviewed at every checkpoint alongside the gap statement. Terms that have been secured move to maintenance; new gaps get new pillars. The core changes slowly. The structure that carries it changes as the record moves.

A typical exclusion list

For a founder, the list often contains the first company, the former job title that still leads the bio, and two or three capabilities that were added to the pitch over the years because a customer asked for them once. For a researcher, it usually contains the institution they left, the earlier field they trained in and the co-author's term that keeps getting attached to their name. For a company, it is the launch product, the geography they left and the industry they no longer serve. None of these are shameful. All of them compete with the core, and every mention of them is a mention the core did not get.

The list is short, usually five to eight items, and it is reviewed with the client before it is applied. The client keeps the final word. The strategist's job is to show the cost of keeping each item, in terms of the associations it dilutes.

Editorial placements come in two forms. Earned coverage is written and published because an editor decided it was worth running. Paid coverage is published because someone paid for the space. Both can carry a semantic core. The difference is not quality; it is disclosure, and disclosure is governed by the rules of the publication and the advertising regulations of the country where it appears.

What earned coverage is

An editor receives a pitch, decides the story matters to their readers and assigns or accepts a piece. No money changes hands for the placement. The piece is independent editorial and is presented as such. Earned coverage carries the most weight with readers and with the sources AI assistants trust, precisely because it was not bought.

It is also the least predictable. Editors decide. A strong pitch to the right title improves the odds; it does not secure acceptance. That is why our commitment for bylines is delivery and submission, and acceptance stays with the outlet.

What paid coverage is

A publication sells space for sponsored content, advertorial or partner features. The content may be written by the client, by the publication's commercial team or by a third party. It appears on the publication's site, sometimes in a dedicated section, and is labeled as paid according to that publication's rules.

Paid coverage is legitimate and useful. It lets you place exactly the claim you want in exactly the outlet you want, on a schedule you control. Because the placement is contractual, our commitment for Publications is defined: publication in an outlet from the agreed list, or a replacement or refund of the unperformed part, as set out in the contract.

Why the label matters

Three reasons. First, the law. Advertising regulators in the UK, the EU and the US require paid content to be identifiable as such. Second, the publication's own trust with its readers, which is the asset you are borrowing. Third, your own record. A piece exposed later as undisclosed sponsorship damages the core more than a hundred properly labeled pieces build it.

There is a further point specific to search and AI. Sponsored sections are often treated differently by indexing systems. A properly labeled sponsored feature in a major title still carries the association; it may carry less weight than earned coverage in the same title. The plan accounts for this by mixing the two rather than relying on either.

How we handle it

Every placement in a plan is marked earned or paid before it is pitched or booked. Paid placements are booked with the label agreed in writing with the publication. Every published piece is logged with its URL, date and label. Publication costs are our own expenses, paid from our account; we never hold client money for placements. The full policy is stated in every contract and on the Commitments page.

How to read a label

Publications use different words for the same thing: sponsored, partner content, promoted, advertorial, in association with, brand voice. All of them mean the placement was paid for. Some titles keep paid content in a separate section with its own design; others run it in the main feed with a label at the top. Some allow the client to write the piece; others insist their commercial team writes it to a brief. Before a placement is booked, we record which words and which format the title uses, and the client sees this before approving the piece. There are no surprises about how the label will look.

Readers are more sophisticated about labels than most clients expect. A well-argued sponsored piece in a respected title is read on its merits. A piece that tries to hide its status is the one that loses trust when it is noticed, and it is always noticed eventually.

A common brief sounds like this: we need more coverage. More is measurable, so it is easy to buy and easy to report. But more is not what search engines and AI assistants reward. They reward consistent associations repeated across independent sources. Twenty pieces that make the same claim with the same terms in the right outlets will move the record further than a hundred pieces that each say something different.

The mechanism

Search systems and language models both learn from co-occurrence. When a name and a term appear together repeatedly, across sources that do not copy each other, the association strengthens. When the name appears with a different term each time, no single association gets strong enough to surface. Volume without consistency produces a wide, shallow record that answers no question well.

This is a mechanism, not a theory. It is why a founder with a modest amount of coverage can be the definitive answer for a category term, and why a company with vast coverage can be described only by its size and funding. The first has a core; the second has volume.

Designing for consistency

Define the core first

Consistency needs something to be consistent with. The Map defines the claims, terms and questions. Everything published afterwards is written to carry them.

Repeat across independent sources

The same claim in the same words on your own site five times counts once. The same claim in a trade title, on a podcast page, in a co-authored piece and in an accurate directory entry counts four times, and each source reinforces the others. Independence is the multiplier.

Vary the form, not the meaning

A feature, a byline, a LinkedIn post and a talk can all carry the same claim in different forms. The reader does not notice the repetition. The systems that index the record do.

What this changes in a plan

The plan becomes narrower and deeper. Fewer outlets, chosen by fit, with several pieces in each. Fewer topics, repeated. Fewer claims, each supported. Reporting changes too: instead of counting mentions, we report core coverage, meaning how many pieces carried each claim and term, and we track the share of voice against named competitors on the same terms.

The objection

Clients sometimes worry that consistency is boring. It is, for the client, who hears the same claim many times. It is not for the reader, who encounters it once or twice in different places and comes away with a clear picture. The clarity is the product. A record that says one thing clearly is understood correctly. A record that says many things loudly is not understood at all.

A number to keep in mind

In our own work, the pieces that move the record most are not the ones in the largest outlets. They are the ones that make the core claim in the outlets the assistants already cite, published in a sequence over several months. A single large feature produces a visible bump in search and a small, temporary change in assistant answers. Six modest pieces in field titles, each carrying the same claim, produce a slower rise that holds. When the two are combined, the large feature becomes the anchor the smaller pieces point back to, and the effect is larger than either alone.

What to ask an agency

If a proposal lists outlets and volumes without stating which associations each piece is meant to build, it is a volume plan. Ask for the core it is meant to serve. If there is no core, the pieces will not add up, whatever their individual quality. This is the question we are happy to be asked ourselves.

Most media lists are ranked by domain authority or audience size. Those measures are useful and we record them. But they answer the wrong question. The question is not how strong the outlet is; it is whether a placement in that outlet will strengthen the associations you want. That is semantic fit, and it comes from the field map, not from a database.

Five tests of fit

Does the outlet already carry the terms?

Search the outlet's archive for the terms in your core. If it has published on them repeatedly, an association between you and those terms in that outlet is credible and will be weighted. If it has never touched the topic, the association is weaker, however large the site.

Do the assistants cite it?

Run the prompts a buyer would ask in your field and note which outlets the assistants cite. Those titles are, by definition, the ones that feed AI answers about your category. A placement there reaches the answer directly. This list is often surprising: specialist titles and association sites appear more than national brands.

Does the buyer read it?

Not the general public. The specific person who decides to hire, fund, cite or partner with you, and the people who advise them. A regional legal publication may reach more of your buyers than a global business title.

Will the piece be indexed and linkable?

Some outlets place sponsored content behind walls or in sections that are not indexed. A piece that cannot be found, cited or linked builds nothing in the record, whatever it does for the reader who sees it that day.

Can the piece make the claim?

Some formats allow a full argument; others allow a mention. A core claim needs space. If the format only allows a quote, the placement may still be useful, but for a term rather than a claim.

Building the list

We run the five tests against every candidate outlet in the field and rank the survivors. The list is shorter than a conventional media list, usually ten to twenty titles rather than a hundred. Each title gets a reason: which terms it carries, which claims it can hold, which buyers it reaches. The reason is what makes the pitch specific and the placement worth having.

Earned and paid on the same list

Fit does not depend on whether the placement is earned or paid. Both are evaluated the same way, and both are labeled correctly when published, as explained in our note on labels. What differs is the commitment: for paid placements we commit to publication in an outlet from the agreed list or a replacement or refund of the unperformed part; for earned, we commit to the pitch and the preparation, and the editor decides.

A worked example

A fintech founder wants to own the term for a specific compliance method. The obvious targets are two national business titles with high authority scores. The five tests tell a different story. Neither title has published on the method; neither is cited by the assistants when asked about it; their readers are general business audiences, not the compliance officers who buy. The titles that pass are a regulatory trade weekly, a specialist podcast with published transcripts and a professional association's journal. All three have modest authority scores. All three are cited in assistant answers about the method. The plan targets those three first, with one national piece later as an anchor once the association exists.

Twelve months on, the founder is named in most assistant answers about the method, and the sources cited are the three specialist titles. The national piece, when it ran, was picked up because the record already supported it.

Most writing for the web is written to be read. That is fine for a newsletter. It is not enough for a piece that is supposed to build a semantic core, because the record is made of citations, links and quotations, not of page visits. A piece that is easy to cite keeps working for years. A piece that is merely pleasant to read is forgotten in a week. The craft of bylined writing in this practice is the craft of being citable.

What makes a piece citable

Clear, quotable claims

A claim is a sentence a journalist or an AI assistant can lift out and attribute. It is specific, complete on its own, and true. Vague sentences are not quoted. Long sentences are not quoted. A paragraph that circles a point three times gives the reader nothing to take. Every core piece should contain three to five sentences that could stand alone with your name attached.

Defined terms

If the piece uses a term from the core, it defines it, briefly, the first time. Definitions get cited. They are what people search for and what assistants extract. Defining a term in a well-placed piece is one of the most effective ways to own it.

Structure that can be skimmed

Headings that state the point, not the topic. Short sections. A callout with the central claim. The person deciding whether to cite you is usually reading fast, and the systems that index the piece read structure as signal.

Sources stated

A claim backed by a named study, a figure or a document is far more likely to be repeated than a bare assertion, and far less likely to be repeated wrongly. Citing your own sources is also what earns citation in return.

The reader is not the only audience

A piece in a good outlet is read by three audiences. The human reader, who wants to learn something. The editor and journalist community, who may reference it. And the indexing and retrieval systems that decide what the piece is about and whether to surface it when someone asks a question. Writing for the third audience does not mean writing badly for the first. It means being clear, which serves everyone.

Where the piece lives

Citability also depends on placement. A piece in an outlet with the right semantic fit, indexed and linkable, carrying a defined claim, will be found and used. The same piece on an unindexed page or behind a wall will not. Writing and placement are one decision, not two.

Reuse

A citable piece becomes the anchor for other work. Its claims become LinkedIn posts. Its definitions become talking points for podcast appearances. Its structure becomes the outline for a talk. One well-built piece can feed a quarter of coverage, all reinforcing the same core, which is the point of consistency over volume.

A checklist before submission

Does the first paragraph state the claim in one sentence a stranger could quote? Is every core term defined the first time it appears? Do the headings state points rather than topics? Is there one sentence in a box that carries the central idea? Are the sources named? Does the piece link to one authoritative page about the author? Would an editor in the field recognise the vocabulary as theirs? If any answer is no, the piece is revised before it goes out. This takes an extra hour per piece and it is the hour that makes the difference between a piece that is read and a piece that is used.

The same checklist is applied to the pieces we write for clients and to the pieces clients write themselves. Writing that passes it is easier to place, because editors recognise a piece that is clear, and easier to measure, because its claims can be found later.

Ask an AI assistant about a professional and watch what it cites. Very often the first source is a LinkedIn profile. The reason is structural. A profile is a labeled, dated, self-declared record of roles, organisations and expertise, in a consistent format the systems know how to read. That makes it one of the most influential documents about you, and one of the least carefully written.

The profile as a structured document

The headline, the about section and the experience entries are read as fields. The headline states what you are. The about section states what you claim. The experience entries state which entities you are linked to and when. Each of these maps directly onto the parts of a semantic core: terms, claims and entities. If the profile does not carry the core, the most-read document about you is working against you.

Common problems. A headline that lists three unrelated roles. An about section written years ago for a different job. Experience entries that name the company but not what you did there or which terms it involved. A profile like this tells the reader nothing consistent, and it tells the systems nothing at all.

Rewriting for the core

Headline

One line. The primary term and the primary entity. Not a list of everything you have done.

About

The central claim, the terms you want to own, the questions you answer, in plain sentences. This is the section most often quoted, so write sentences that can be quoted.

Experience

Each entry links you to an entity and states, briefly, the terms and claims associated with that role. This is where associations between your name and organisations are made explicit.

Posts reinforce; they do not replace

Posts matter, but differently. A steady sequence of posts structured by the core, published from your own account with your approval, reinforces the associations the profile declares. That is what LinkedIn ghostwriting does in this practice: every post carries a defined part of the core, and the monthly plan covers the whole map over a quarter.

What posts do not do is substitute for the profile. Posts are transient; the profile is the standing record. A great post sequence on top of an outdated profile still leaves the assistant citing the outdated profile.

What we do not do

We do not run your account, use automation that violates platform rules, or buy interaction of any kind. We do not report audience or interaction figures as outcomes, because they are not what the work is for. We report the posts delivered and the parts of the core they covered, and the Presence report shows whether the record moved.

A before and after

Before: a headline reading Founder, Advisor, Speaker, Investor. An about section from 2021 describing a company that was since sold. Experience entries that name employers and dates and nothing else. Assistant answers describe the person by the sold company and the word advisor. After: a headline naming the current company and the category term. An about section with the central claim, three defined terms and the questions the person answers. Experience entries that state, in one line each, what the role involved and which terms it carried. Three months later, assistant answers name the current company and the category term in most runs, and the profile is among the sources cited.

Nothing about the after version is clever. It is the same information, stated clearly, in the fields the systems read. Most profiles have never been written that way because nobody told their owner that a machine was the main reader.

A podcast appearance is usually judged by the conversation: was it a good interview, did the host ask the right things, did it reach many listeners. Those matter. But the durable value of an appearance is in the episode page. The title, the description, the guest bio, the transcript and the links are what search engines index and what AI assistants read when they answer questions about you. The conversation is heard once; the page is read for years.

What the page carries

The title

If the title names you and the core term, the association is made in the strongest possible position. If it names only the host's theme, the association is buried in the transcript.

The description

Usually two hundred words, usually written by the producer from the recording. This is the paragraph the assistants quote. It should state who you are, which entity you belong to, and the claim the episode makes.

The transcript

More shows publish transcripts every year, and the systems read them in full. A transcript in which you define the core terms clearly and make the core claims in quotable sentences is a long, indexed, independent source about you.

The links

A link to your site or profile from the episode page connects the two records. It also tells the systems which page is the authoritative source for your name.

Choosing shows by field

The same logic as choosing publications applies. A show whose hosts already discuss your terms, whose past guests share your semantic field and whose episode pages are indexed and cited is worth more than a larger show outside the field. Download counts are recorded; they do not decide.

Preparation

Preparation is where most of the value is created. A prep pack with the host's style, likely questions, your three core points, the terms to use and the ones to avoid means that the sentences you want quoted actually get said, clearly, in the recording. This is the podcast service in one sentence: the right shows, prepared properly, followed through.

After release

Log the episode with URL and date. Share it with the outlets and contacts in your field. Cut approved clips and quotes for LinkedIn. One appearance becomes several citable pieces carrying the same associations, and each one links back to the episode page, which strengthens it further.

What we do not promise

Hosts decide their own guests. We commit to pitching an agreed list and preparing you fully. We do not promise a booking by any specific show, and any commitment stated otherwise would be misleading.

A note on transcripts

Transcripts are the fastest-growing source about individuals in AI answers, because they are long, specific and written in the guest's own words. A guest who defines a term clearly in an interview has produced a definition that may be quoted for years. A guest who rambles has produced a long page that says nothing the systems can extract. This is the strongest argument for preparation: not to sound polished, but to make sure the sentences that matter get said in a form that can be lifted out. We ask clients to say the core claim once in full, early in the conversation, and to use the core terms rather than synonyms.

After release, we read the transcript and note which core claims and terms appear in it. That note goes in the delivery log alongside the episode URL, and it feeds the next month's measurement.

Any practice that claims to influence how AI assistants describe a client has to answer a hard question: how do you measure it honestly? The answers vary between sessions, between models, between phrasings and over time. A single screenshot proves nothing. What is needed is a method that produces a fair before-and-after comparison, states its limits and can be checked. This is ours.

Principles

Fixed prompt set

Before the work starts we agree a set of prompts: who is X, what is X known for, what does X's company do, and the five to ten questions a buyer in the field would actually ask. The set is written down and does not change between checkpoints, so the comparison is like for like.

Repeated runs

Each prompt is put to each assistant several times in separate sessions. We record every answer. What we report is the pattern: how often a term appears, how often an entity is named, which sources are cited, where the answer is wrong. One run is an anecdote; a pattern is data.

Multiple assistants

We test the leading assistants, not one. They draw on different sources and behave differently. A change that appears in all of them is a change in the record; a change in one may be noise.

Sources logged

Where an assistant cites sources, we record them. This is the most useful part of the measurement because it shows which pages the answer came from, and therefore which pages need to carry the core.

The baseline

The first run is the baseline, taken in the Audit before any coverage. It is reported alongside the search baseline and the gap statement. Everything later is compared with it.

Checkpoints

Checkpoints are agreed in advance: typically at three, six and twelve months for a campaign, and monthly for Presence subscribers. The same prompts, the same assistants, the same number of runs. The report shows the baseline and the checkpoint side by side, term by term, with the sources cited in each.

What the report does not say

It does not say that coverage caused a change, because that cannot be proved from the outside; it says that the record changed and the coverage was published in between. It does not report a single answer as the answer. It does not claim that any assistant will name the client. Those limits are stated in the report and in the contract, because a measurement that overstates itself is worse than no measurement.

Why we publish the method

So that a client, a board or an agency can check it. Anyone can run the prompt set themselves. If our reports and theirs disagree, we want to know. A method that can be checked is the only kind worth paying for.

The report itself

The report is short. One page per assistant, one table per prompt. Each table shows the baseline and the checkpoint side by side: which terms appeared and how often, which entities were named, which sources were cited, where the answer was wrong. A summary page lists the associations that strengthened, the ones that did not move and the ones that weakened. There is no commentary about why, because the why is inference and the report is data. The client, their board or their agency can read it in ten minutes and check any line by running the prompt themselves.

We have found that clients trust this format more than a narrative, precisely because it does not argue. It shows the record before and after, states its limits and stops.

Share of voice is an old measure in media relations: what fraction of coverage on a topic mentions you rather than your competitors. The idea still works, but the topic and the coverage have changed. The topic is now a set of terms from your semantic core. The coverage is now search results, AI answers and trade media together. Measured this way, share of voice becomes the clearest single view of whether the record is moving in your favour.

Defining the comparison

Share of voice needs three inputs agreed in advance. The terms: five to ten from the core. The competitors: three to five named people or companies you want to be compared with, chosen from the field map. The channels: search results for the terms, AI answers to the prompt set, and mentions in the agreed list of trade outlets. Change any of the three later and the series breaks, so they are fixed at the start.

How it is measured

Search

For each term, the top results are recorded and each is scored for which of the named parties it associates with the term. Your share is your associations divided by the total.

AI answers

For each prompt, across repeated runs and several assistants, the answers are scored for which parties are named and how prominently. Because answers vary, the score is a proportion across runs, not a single result.

Trade media

For the agreed outlets, mentions of each party in connection with each term are counted over the period. Volume is recorded, but the score weights pieces that carry the term, not pieces that merely mention the name.

Reading the numbers

A rising share on an open term is the expected result of a well-placed campaign. A flat share on a contested term tells you the owner is still stronger and the plan should narrow or partner. A falling share, with the same coverage, usually means a competitor has started publishing on the term, which the field map should catch at the next checkpoint.

The numbers also expose mismatches between channels. A strong share in trade media and a weak share in AI answers usually means the outlets carrying you are not the ones the assistants cite. That is a placement problem with a placement solution, explained in choosing by fit.

In the monthly report

Presence subscribers receive share of voice every month alongside the search and AI tracking. It is presented as a simple table: term by term, party by party, channel by channel, this month against last month and against the baseline. No commentary is needed to read it, which is the point.

What it is not

It is not a measure of reach, audience size or traffic, and it does not report those. It is not a ranking, and it does not claim one. It is a proportion of the record, for the terms you chose, against the parties you chose. Within those limits it is honest and useful, and outside them it would be neither.

A worked example

A specialist law firm chooses five terms and three competitor firms. At baseline its share on the terms is small in search, near zero in assistant answers and moderate in the two trade titles that matter. The plan targets the assistant gap first, through the titles the assistants cite and through accurate directory entries, which the competitors have neglected. Six months later the search share is up modestly, the trade share is roughly unchanged and the assistant share has risen sharply, because the firm is now among the few sources the assistants find for those terms. The table shows this without commentary, and the next plan targets the trade titles, where the competitors are still stronger.

The example is ordinary and that is the point. Share of voice is not a headline number. It is a working tool that tells the plan where to go next.

Sooner or later everyone finds something wrong about themselves online. An old title. A former employer listed as current. A project attributed to the wrong person. A figure misquoted in an article. These things are correctable, through channels that exist for the purpose. What is not correctable is accurate information you would prefer did not exist. Any practice that blurs this line is selling something it cannot deliver.

What can be corrected

Factual errors in articles

Publishers have corrections policies. A request that identifies the error, provides the correct information with evidence and follows the publisher's process is usually granted, and the correction is noted on the piece. This is the core of Accuracy and Correction.

Outdated profiles and directories

Many of the sources AI assistants cite are profiles and directories that can simply be updated, either directly or by request. This is unglamorous work and it is often the highest-value correction available, because those sources are copied by others. The sources nobody checks explains why.

Personal data, where the law applies

Data-protection law in the UK and the EU gives individuals rights to rectification and, in defined circumstances, erasure of their personal data. Where the right exists, a request with the correct legal basis can be sent to the controller. Where it does not exist, no request should be sent, and we say so.

Matters for a lawyer

Defamation, breach of contract, regulatory misstatements. These need a lawyer. We prepare the factual record and coordinate with the lawyer you appoint. We do not give legal advice.

What cannot be corrected

Accurate reporting you dislike. A review that is unfavourable but genuine. A public record of something that happened. Publishers and platforms make their own decisions about such content and they are under no obligation to change it. A request to do so is a request they will refuse, and repeated requests damage your standing with the editor. We do not make them.

There is also a category of claims to avoid entirely: promises to push content down, to control what search returns, or to make a name appear at the top of results. No one can deliver those honestly. A firm that offers them is either misleading you or planning to do something that will hurt the record later.

The better route

The most effective response to accurate but unwelcome content is a stronger, more current, more consistent record around it. A well-published semantic core gives readers and systems more accurate material to draw on, and the older material becomes one source among many rather than the only one. That is coverage, not correction, and it is built the same way as any campaign.

How requests are handled

Every request is logged with its basis, the date sent, the evidence attached and the outcome. You see the log. Requests are followed up on the agreed schedule. Where a request is refused, the refusal and its reason are recorded. The commitment is requests sent and tracked; the decision belongs to the publisher, the platform or the controller.

A short example

A physician finds that a widely-used medical directory lists her at a hospital she left four years ago, with a specialism she no longer practises. Three other directories have copied it. Two assistants describe her from it. The correction is a request to the source directory with her current registration details, followed by requests to the three copies once the source is fixed, then a check of the assistant answers a month later. The whole exercise takes six weeks, most of it waiting. It changes the assistant answers more than any article would have, because it fixes the page the assistants were reading.

Nothing in this example involved a lawyer, a takedown or a dispute. It was a factual error, corrected through the channel that exists for it. Most corrections are like this, and the ones that are not are handed to a lawyer.

When we trace where an AI assistant got a wrong fact about a client, the answer is rarely a major publication. It is usually a directory. A company register entry, a professional association's member page, a conference bio from four years ago, a data vendor's profile assembled from other directories. These pages are dull, rarely updated and copied endlessly. The systems trust them because they are structured, and because they agree with each other, even when they are all wrong together.

Why they carry so much weight

Three reasons. They are structured: name, organisation, role, location, in labeled fields the systems read easily. They are numerous: a single wrong entry is copied into a dozen others within months. And they are stable: they do not change, so they accumulate links and citations over years. Together this makes a directory entry one of the most-weighted sources about a person or a company, out of all proportion to how often a human reads it.

The usual errors

An old role listed as current. A former company as the present one. A city you left. A phone number that rings a desk that no longer exists. A specialism that was accurate in 2019. A misspelled name that splits your record in two. None of these is dramatic; all of them are repeated by assistants with complete confidence.

The audit

The Audit includes a source trace: for every association the assistants and search results attach to you, which page it comes from. Directories appear on that list more often than any other source type. The trace tells you which entries need updating first, ranked by how many other sources copy them.

The fix

Most entries can be updated directly or by a request to the operator. The work is unglamorous and it is done in Accuracy and Correction: find the entry, record the error, update or request, confirm the change, log the date. Where an entry copies from another entry, fix the source first and the copies follow. For Presence subscribers this is checked every month, because new entries appear and old ones get re-copied.

Using them well

Directories are also an opportunity. A correct, current entry that carries your core terms and links to your site is a strong, stable source in your favour, for the same reasons a wrong one is a strong source against you. Part of building a core is making sure the dull pages say the right thing. It is the cheapest coverage you will ever place, and among the most durable.

Which directories matter

It depends on the field. For a company: the official register, the main business data vendors, the trade association. For a professional: the regulator's register if there is one, the professional body, the major conference archives, the institutional page. For a researcher: the institutional profile, the publication databases, the funder's records. The field map lists the ones that the assistants actually cite in your category, which is the list that matters.

A short checklist

Search your own name and your company name with the word directory, profile and register added. Open every result on the first two pages. For each, note whether the entry exists, whether it is correct, whether it links to your site and whether it uses your core terms. Fix what you can directly. List what needs a request. Repeat in six months. This takes an afternoon for most people and it is among the highest-value afternoons available in this work, because the entries you fix are the ones that get copied next.

Clients are sometimes surprised that a semantics practice spends time on something so mundane. The answer is that meaning is made from sources, and these are the sources. The dull pages are where the record lives.

Marketing teams brief agencies. Agencies brief writers. Writers produce pieces. At each hand-off, some of the meaning is lost, and by the time the piece is published it says something slightly different from what the founder would have said. Multiply by a year of coverage and the record is inconsistent. A semantic core, in the form of the Map, is the document that stops the drift, because every party can be briefed against it and every piece can be checked against it.

For in-house teams

The Map gives an in-house team three things it usually lacks. A short, agreed statement of what the company should mean, signed off by the founder. A filter for saying no to opportunities that do not fit. And a checking standard for everything an agency delivers. The last is the one teams value most: instead of judging a draft on taste, the team judges it on whether it carries the pillar, the claim and the terms.

It also survives staff changes. A new marketing lead inherits the core and the architecture rather than a folder of past campaigns and a vague sense of what worked. The structure and filter are written to be handed over.

For PR firms

Agencies that already run placements can add the semantic layer without building it themselves. Under a white-label arrangement we deliver the Audit and the Map to the agency, in the agency's format if preferred, and the agency briefs its own writers and pitches its own placements against them. We invoice the agency only and never contact or collect payment from the agency's end clients.

The benefit for the agency is a plan the client cannot argue with. Instead of a media list, the agency presents a gap statement and a core, and every recommended placement has a reason attached: this outlet carries this term, this piece makes this claim. Retention improves because the reporting is about the record moving, not about counts.

The brief template

We supply a one-page brief template with the Map. It states the pillar, the primary claim, the supporting claims and evidence, the terms to use, the terms to avoid, the question the piece answers and the sentences we want quoted. A writer who follows it produces a citable piece on the first draft. An editor who reads it knows exactly what is being offered.

Checking the output

Once a quarter, take ten published pieces and score them against the core. Eight or more carrying the pillar, the claim and the terms means the brief is working. Fewer means the filter is being bypassed or the brief is being ignored, and either is fixable. The same check feeds the share of voice measurement, so the team can see the connection between discipline in briefing and movement in the record.

What changes

Less coverage, more consistent. Fewer arguments about taste. Reporting that a board can read. And a public record that says one clear thing, which is what a core is for.

A note on ownership

When an agency uses the Map under a white-label arrangement, the core belongs to the end client and the agency holds it on their behalf. If the client changes agencies, the core goes with them, and the new agency can be briefed against it on day one. This is deliberate. A core that belonged to the agency would be a lock-in device, and lock-in is the opposite of what a clear public record needs. Agencies that work this way find that clients stay because the work is good, not because leaving is hard.

In-house teams own their core outright and may share it with any partner they choose. We recommend sharing it widely: with the agency, the writers, the event team, the person who updates the directory entries. The more people who brief against the same document, the more consistent the record becomes.