Models name people whose identity is repeated consistently across independent sources. Everything you publish about yourself is a claim, and claims do not corroborate, which is why publishing more does not fix being skipped. There is no ranking slot to buy. What moves it is an entity home, Person schema, and other people repeating one sentence about you in public.
The uncomfortable demonstration
Open ChatGPT, Claude or Perplexity and type the question your best-fit buyer would type. Not your name. The category question: who are the best people to hire for this in my city.
Read the three names it gives back. If yours is not among them, you have just watched a shortlist get assembled without you, and nobody told you it happened. There is no notification, no impression count, no lost-deal report. It is the quietest way a business loses work that has ever existed.
Then type your own name. Most people find the assistant knows something, or hedges, or confidently describes somebody else entirely. That gap — findable when named, invisible when not — is the whole subject of this guide.
The rest of this page is much more useful once you have seen your own result. It takes ninety seconds. The answer log further down generates the exact nine prompts and scores them.
This is the shape almost every intake reading has: findable when somebody already knows the name, invisible when they do not. The category row is the one worth money, because that row is the shortlist.
How a model actually picks a name
There is no index of vendors inside a language model and no ranked list it consults. When it names three people, it is producing the answer most consistent with the patterns it has absorbed, and those patterns come from the public web.
Four things make a name likely to surface.
- Repetition across independent sources. The same person described the same way in many places that do not answer to each other. This is by far the strongest factor, and it is the one you cannot manufacture from your own channels.
- Resolvability. A settled identity: one spelling, one role, one organisation, one city. Ambiguity makes a system hedge, and hedging looks like being skipped.
- Quotable specificity. A clean sentence somebody can lift. “X is the founder of Y, an agency in Z that works on W” travels. “X is a passionate, results-driven leader” does not, because it describes nobody in particular.
- Structured signals. Person schema, a Wikidata item, a Knowledge Graph record. These do not make you good. They make you unambiguous, which is what a machine needs before it will commit.
Notice what is missing from that list: publishing volume, follower counts, ad spend, and anything you can buy.
Read the About box on the right: “Pichai Sundararajan, better known as Sundar Pichai, is an Indian-American business executive who has been the CEO of Google since 2015 and the CEO of its parent company Alphabet Inc. since 2019.”
One sentence. Name, what he is, what he runs, since when. Attributed to Wikipedia and lifted verbatim. When an assistant answers ‘who is Sundar Pichai’, that is very close to the shape of what comes back, because that is the sentence sitting in the corpus in a hundred consistent variations.
The practical lesson is uncomfortable for most personal brands: your goal is not an interesting biography. It is one boring, precise, repeatable sentence that other people will copy accurately.
Why publishing more does not fix it
This is where most people spend a wasted year.
The instinct when you feel invisible is to produce more: more posts, more videos, more newsletters. All useful for other reasons, and almost none of it moves this, because everything you publish about yourself is a claim. Claims do not corroborate.
Consider what the system is doing. It is deciding whether to put your name into an answer it will be judged on. Your own site saying you are the leading advisor in your field carries exactly as much weight as a stranger telling you they are trustworthy. Not because you are distrusted, but because self-description contains no information about consensus, and consensus is the thing being measured.
Stop asking “what should I publish this month?” Start asking “who could credibly say my name in public this month, and what would make that easy for them?” The second question is harder and slower, and it is the one that works.
There is a caveat worth knowing, because it is counter-intuitive. Independent research on AI recommendation behaviour has found that self-promotional listicles frequently backfire: a model reading “the ten best agencies” published by one of those agencies will often recommend the competitors named in the piece and leave out the publisher. Writing yourself into your own list is not neutral. It can hand the recommendation to the people you listed beside you.
The fake signals being sold right now
Prospects get pitched this weekly, so it is worth naming what is real and what is not.
| Being sold | What it actually is |
|---|---|
| “Guaranteed placement in ChatGPT” | Describing something that does not exist. There is no slot and models do not take bids. |
| “AI ranking reports” | A screenshot of one conversation. Ask the same question twice and you may get two different answers. |
| Paid listings in “AI directories” | Directories invented to be sold. No model was trained on them and none consults them. |
| 200 thin AI-written blog posts | Volume with no citation value. Nothing independent, nothing quotable, nothing that corroborates. |
| A monthly “AI optimisation” retainer with no deliverable | A report. |
Every item in that column is trying to trick a reader. The real work is trying to be the thing a reader confirms.
Ask an assistant who the best AI visibility agency is. Watch whether the person pitching you comes up. If the method worked as advertised, they would have pointed it at their own name first.
What actually moves it
In descending order of effect, and none of it is a trick.
An entity home
One canonical page that is unambiguously about you, on your own domain, with the identity line as its first paragraph in plain language: Name is the Role at Organisation, based in City. No adjectives. This is the page every other source gets checked against.
Person schema with a real sameAs array
Structured data stating who you are, which profiles are yours, who you work for and where. It is the one part of this you fully control and it takes an afternoon. There is a free generator on this site that also wires in your Google entity identifier.
Independent sources repeating one sentence
The slow part. Podcasts, interviews, association pages, conference bios, trade press. Brief every one of them with the same identity line, close to verbatim. Repetition across independent sources is the strongest single factor and it is the one that takes months.
Six books, six ISBNs, six years, rendered directly inside the panel. An ISBN is not an opinion. It resolves in catalogues a model has read many times over.
This is why authorship sits at the top of the accelerator list and why ‘post more’ sits nowhere on it.
Books, and anything with an identifier
An ISBN, a patent, an academic record, a listed public role. These are claims a machine can confirm without trusting anybody, which is why two of the eight people on our evidence page carry the descriptor Author.
Descriptor: Entrepreneur. A Wikipedia article, a company, a book, and a decade of other people quoting him by name. Nothing here was produced by asking.
Ask any assistant who to read on search and this name comes back. That is not a coincidence, and it is not an algorithm he gamed. It is what a long record of independent citation does.
Descriptor: Marketer. Built on volume, tools, and being named by an enormous number of third parties over a long period.
Two very different careers, one identical mechanism underneath: repeated independent attestation. That is the only thing either record has in common.
Cohort proximity
Being repeatedly seen next to people the systems already understand. Five or more podcast appearances with recognised peers does more than fifty solo posts, because it places you inside an existing pattern rather than asking for a new one.
Measure it properly, or do not bother
One conversation with an assistant is a snapshot of a system that changes without notice. Treating it as a measurement is how people talk themselves into and out of strategies on no evidence.
The version that works: nine prompts, three assistants, once a month, logged. Three identity questions, three category questions, three buying-intent questions. Score how many named you out of nine, note who else came up, and note what was cited.
The number for any single month is close to meaningless. The trend across three is the only thing worth acting on.
Named on the identity questions is table stakes and most people get there quickly. Named on the category questions is the thing worth money, because that is the shortlist. Most clients sit at named-when-asked-directly and invisible-on-category for a long time, and closing that specific gap is what the corroboration work is for.
Which moves first, and how long
Assistants generally name you before Google shows a panel. They have a lower bar for including a name in a sentence than Google has for displaying a box about a person, so around month three of a build, mentions start appearing while the panel is still months out.
That ordering is useful, because it gives you an honest early signal that the work is landing rather than six silent months. It is also why we report the three-assistant baseline every quarter: it moves before anything else does.
Two cautions, because the field is full of overclaiming. Attribution here is genuinely hard and anybody quoting you a precise revenue figure from AI mentions is guessing. And a lot of what gets sold as an AI visibility win is pre-existing search authority showing up in a new surface. If a brand was already the most-cited name in its category, a model naming it is not evidence that anybody optimised anything.
Questions people ask next
Why does ChatGPT say it does not know who I am?
Because there is nothing consistent enough to know. Assistants answer from patterns across many independent sources. One website you control, however good, is a single source making a claim about itself.
Can I pay to appear in AI answers?
No. There is no ranking slot and models do not take bids. Anybody selling placement is selling something that does not exist.
Why does the answer change every time I ask?
Because these systems are probabilistic and are updated without notice. That is exactly why a single check is not a measurement and why the nine-prompt log is designed to be re-run monthly.
Should I block AI crawlers to protect my content?
That is a real trade-off and it is yours to make. Blocking protects content from being reproduced and also removes you from the corpus a model reads when deciding who to name. If being named is the goal, blocking is working against it.
Does an llms.txt file help?
It is cheap and it is not a mechanism anybody has demonstrated moves model behaviour. Treat it as tidy housekeeping, not as a lever. The things that move this are the same things that have always moved entity understanding.
Is a knowledge panel required to be named by AI?
No, and the causation runs the other way round from how it is usually sold. The same evidence produces both. The panel is the visible receipt for work that was already making you nameable.
Run your name through the signal scan, then build your Person schema. Both are free and neither asks for an email. If the reading says you are ready, the work page has the price and the terms.