The short answer
If you read nothing else, read this part.
Gemini grounds its answers in Google Search results and the Knowledge Graph, so getting named comes down to three things: being resolvable as an entity (one consistent name, niche and role everywhere you appear), publishing short standalone answers to the decision-shaped questions your clients actually ask, and being corroborated by sources that are not you.
The third one is the one almost everybody skips, and it is the one that does the heavy lifting. A claim that appears only on your own website is a claim with one source. A claim that appears on your site, in a podcast episode description, in a professional directory and in a client's write-up is a claim with four. Language models are built to prefer the second kind, and no amount of on-page work substitutes for it.
One caveat before the detail, because most articles on this subject will not give you one: Google has never published how Gemini selects sources, and it is not obliged to. Everything below is either a documented behaviour, a controlled study, or an observation from our own data, and it is labelled as such. Treat anyone who claims certainty here with suspicion.
How Gemini picks who to name
Three mechanisms sit between your website and a sentence in an AI answer.
Grounding
When a question needs current or specific information, the model runs searches and answers from what comes back. Pages that Google has not indexed cannot be used, which makes indexation a hard prerequisite rather than a nice-to-have.
Query fan-out
Google has said its AI Mode breaks one question into several related sub-queries and searches each. This is why you get named for questions you never targeted, and why a page answering one narrow thing well can surface under a much broader prompt.
Entity resolution
The model has to work out that the "Sarah Chen" on your site, the one on the podcast and the one in the directory are all the same person. When it cannot, your evidence is split across three weak entities instead of one strong one.
Why this is different for a coach than for a software company
Most advice on this topic is written for SaaS companies, and it quietly assumes two things that are not true of a coaching practice. It assumes the entity is a company, and it assumes third-party corroboration is easy to get because software has review sites.
For a coach, the entity is usually a person, which is harder to resolve and much easier to fragment. And there is no G2 for coaching. The corroboration exists, but it is in different places: professional body listings, podcast guest appearances, conference and workshop billings, published client outcomes, press mentions, and co-authored work. Those are your review sites. They are slower to accumulate and they are worth more.
What does not work
Both of these have been tested against controls. Both came back empty.
Schema markup
Ahrefs, Aug 2025 to Mar 20261,885 pages that added JSON-LD, measured against roughly 4,000 matched controls. Null effect on ChatGPT citations, null on Google AI Mode, and a small negative effect on AI Overviews. No major AI crawler is known to read it. Keep schema, because it earns rich results in classic search and that is what it was built for. Just stop counting it as AI work.
llms.txt
SE Ranking, May 2026Around 300,000 domains studied. Roughly 10 percent adoption, no confirmed support from any major AI vendor, and near-zero presence in the content engines actually cite. It costs ten minutes to add and it does no harm, so add it if you like. It is not a strategy, and anyone selling it as one is selling you a text file.
Chasing the big keywords
Our own dataWe hold dedicated, substantial pages for the largest terms in our category and they rank in the forties. Meanwhile our single best-performing page targets a question with a keyword difficulty of zero, ranks first, and drives 81 percent of our tracked organic traffic. It is also the shortest page we have. Length was never the lever. Difficulty was.
The pattern
Every tactic that failed was a file you can add in an afternoon. Every tactic that worked took months and involved other people.
What actually works
Four factors account for roughly 70 percent of the variance in whether a source gets cited.
| Factor | What it means for a coach | Effort |
|---|---|---|
| Source authority | How established your domain and your name are. The slowest to move and the hardest to fake. | Months |
| Answer concision | Whether a passage can be lifted and quoted without the paragraphs around it. | Hours per page |
| Cross-source consensus | Whether the same claim about you appears on sites you do not control. | Months, ongoing |
| Parseability | Whether the text is readable without running JavaScript, behind no login, no interstitial. | One-off fix |
Note the fourth row, because it is the one people get wrong in a way that is invisible to them. If your site renders its content with client-side JavaScript and serves an empty shell to anything that does not execute it, a good number of AI crawlers see nothing at all. You will not notice, because your browser runs the JavaScript perfectly. The check takes one command:
curl -s https://yoursite.com/your-page | grep "a distinctive sentence from the page"If that returns nothing, your content does not exist as far as a non-rendering crawler is concerned, and no amount of writing will fix it. That is a build problem, and it outranks everything else on this page.
What earns citations that generic content does not
Original numbers. Anything counted from your own practice that nobody else can count. Client outcomes across a cohort, session data, a survey of your list. Original figures get quoted because there is no alternative source for them.
Named case studies. A real client, a real starting position, a real result, with their permission and their name on it. The specificity is the point. Anonymised composites read as marketing and get treated as such.
Honest comparisons. Where your approach fits against the alternatives, including where it does not fit. Comparison content is disproportionately cited because it is what people are asking about at the point of decision.
The unfashionable answer. A defensible position that contradicts the consensus, with the reasoning shown. Models tend to surface the disagreement when one exists, and there is far less competition for it.
The seven-step playbook
In order. The early steps are cheap and the later ones are not, so do not skip ahead.
Confirm you are actually indexed
Check Google Search Console and Bing Webmaster Tools coverage reports directly. Do not use a site: search to judge this, on either engine. The operator undercounts badly and we have watched it send people chasing an indexing problem that did not exist.
Effort: 20 minutes · Blocks everything else
Fix one name, one role, one niche
Pick the exact string you want associated with you and use it identically on your site, LinkedIn, every directory listing, every podcast bio and every speaker page. "Executive coach" on one and "leadership consultant" on another splits one entity into two weak ones.
Effort: an afternoon · Highest ratio of result to work
Write down the twenty questions you are actually asked
Not keywords. The literal sentences prospects say on discovery calls, in DMs and in your inbox. These are already phrased the way people phrase things to an assistant, which is the whole point. You are not researching this list, you are transcribing it.
Effort: one hour if you have been paying attention
Answer each one in a passage that stands alone
Lead with the direct answer in the first two sentences, then support it. The test is whether a paragraph still makes sense pulled out of the page with nothing around it. If it needs the previous section to be intelligible, it cannot be quoted.
Effort: ongoing · The core of the work
Publish something only you can publish
Your numbers, your cohort results, your survey of your own audience. One piece of genuine original research outperforms a year of restated general advice, because restated advice has no reason to be cited over its source.
Effort: weeks · Highest ceiling
Get corroborated somewhere you do not control
Professional body listings, podcast guest spots, industry directories, conference billings, client write-ups, co-authored articles. Each independent source that repeats the same facts about you strengthens every claim on your own site.
Effort: months, ongoing · The real constraint
Measure, then decide whether to keep going
Set up the two signals in the next section before you invest further. Plenty of coaching niches have audiences that simply do not use assistants to find practitioners, and finding that out after two months of work is a great deal cheaper than finding it out after a year.
Effort: an hour to set up · Do this before scaling spend
How to measure it
Your analytics will show almost none of this. Two other signals will.
The reason AI visibility feels unmeasurable is that a citation usually produces no click. Someone asks Gemini or ChatGPT a question, gets an answer that names you, and never visits your site. The value is real and your analytics records nothing. Two sources close most of that gap.
1. Bing Webmaster Tools, AI Performance
This is the single most underused free tool in this whole area. It reports observed citation counts, not a modelled estimate, along with the grounding queries that produced them. It covers Copilot and partners, and because ChatGPT's browsing retrieves from the Bing index, it is also the closest proxy you have for that channel. It is free, it takes ten minutes to verify a site, and it will tell you what questions you are already the answer to.
Worth knowing what a gap between tools looks like. Over one recent three-month window, Bing reported around 1,900 real citations for our own site while a well-known paid SEO suite reported an AI visibility score of zero for the same period. The paid score is a sampled estimate and it under-detects heavily at modest traffic levels. If a dashboard tells you that you have no AI visibility, check the observed data before you believe it.
Observed citations in 3 months, Bing AI Performance
Reported by a paid SEO suite over the same window
Cost of the tool that was right
2. The zero-click pattern in Search Console
Filter Search Console for long, conversational queries. Full sentences, the way somebody talks to an assistant rather than the way they type into a search box. Then look at the ones ranking in positions one to five with a click-through rate of zero.
The instinct is to read that as a failure, and it usually is not. Ranking first and getting no clicks on a sentence-shaped query is close to a signature for an assistant retrieving the page mid-conversation. We have dozens of these and they are among the most valuable positions we hold. Judge them on citation, not clicks.
And add one question to your intake form: "Where did you hear about us?" with an AI assistant as an explicit option. It is the only attribution that survives a zero-click channel, and it costs one form field. In our own sample of 25 signups, 44 percent named ChatGPT or another assistant. That is a small sample from one product, which is exactly why you should be collecting your own.
Where an AI clone fits
An honest account, including the part that will not help you.
We build AI clones for coaches, so treat this section with the scepticism it deserves. Here is the accurate version rather than the flattering one.
What it will not do: having a clone does not make Gemini cite you. There is no mechanism by which it would. Anyone implying otherwise is guessing.
What it does do is generate the raw material for step three and step five of the playbook above. Every question a client asks your clone is a real question, phrased the way real people phrase things, logged and countable. That is the transcription exercise in step three, done automatically and at a volume you could not reach by memory. It is also original data of the kind described in step five: we published an analysis of 4,670 questions across 1,810 conversations for exactly that reason, and it is a piece nobody without that data could have written.
A public clone is also a page a person can interact with rather than read, which tends to be linked and talked about more than a static page on the same subject. Ask David, built on the work of coaching author David Clutterbuck, is a live example you can try without signing up for anything.
See your AI clone in 30 seconds
Paste your website. Watch it learn what you do, who you help, and how you sound.
Free, no signup needed to preview.
Realistic timelines
What to expect, and when to conclude it is not working.
Indexation confirmed, entity names unified, question list written. Nothing visible has happened yet and nothing should have.
Individual long-tail questions start returning your name. Bing AI Performance registers its first grounding queries. Volumes are small and that is normal.
Category-level questions become reachable, but only if step six is genuinely underway. If you skipped the corroboration work, this is where progress stops and does not resume.
Either the trend in observed citations is climbing or it is flat. Flat after six months of real effort is a legitimate answer: your audience may not be finding coaches this way. Believe your data.
One number for context on why this is worth the patience: Gartner has predicted that traditional search engine volume will fall by 25 percent by 2026 as people move to assistants. Predictions of this kind are frequently wrong about the timing and directionally right about the shift. The safe reading is not that classic search is finished, but that a growing share of the people looking for a coach will never see a page of blue links while deciding.
Frequently asked questions
The questions that come up most often on this subject.
How do I get my coaching practice into Google Gemini answers?+
Gemini grounds its answers in Google Search results and the Knowledge Graph, so the practical route is threefold. First, be resolvable as an entity: use one consistent name, niche and role across your site, LinkedIn, directory listings and podcast bios. Second, publish answers to the decision-shaped questions your clients actually ask, in short standalone passages that can be quoted without surrounding context. Third, get corroborated by independent sources, because a claim repeated across several sites is far more likely to be used than one that appears only on your own.
Does schema markup help you get cited by AI?+
No, on the current evidence. Ahrefs ran a causal study from August 2025 to March 2026 comparing 1,885 pages that added JSON-LD against roughly 4,000 matched control pages. The effect on ChatGPT citations was null, the effect on Google AI Mode was null, and the effect on AI Overviews was slightly negative. No major AI crawler is known to read it. Keep schema for Google rich results, which is what it was designed for, and expect nothing else from it.
Is llms.txt worth adding to a coaching website?+
It is close to free to add and close to worthless as a ranking lever. SE Ranking studied around 300,000 domains in May 2026 and found roughly 10 percent adoption, no confirmed support from any major AI vendor, and almost no presence of llms.txt files in the content that engines actually cite. Add it if it takes ten minutes. Do not build a strategy on it.
How long does it take to appear in AI answers?+
A new page can be picked up within days once it is indexed, but that is the exception rather than the plan. For a coaching practice starting from low authority, expect one to two months before individual long-tail questions start returning your name, and closer to six months before you appear reliably for the broader category questions. The limiting factor is almost never the page itself. It is how many independent sources corroborate what the page says.
How can I tell if AI assistants are citing me?+
Google Analytics will not tell you, because most citations produce no click. Two reliable signals exist. Bing Webmaster Tools has an AI Performance panel reporting observed citation counts and the grounding queries behind them, which covers Copilot and ChatGPT search. And in Google Search Console, long conversational queries ranking in positions one to five with a zero percent click-through rate are usually an assistant reading the page rather than a human ignoring it.
Is AI search visibility actually worth pursuing for a coach?+
It depends entirely on your audience, so measure before you invest. As one data point, in a sample of 25 recent Personify signups asked where they heard about us, 44 percent named ChatGPT or another AI assistant and 36 percent named Google search. That is a small sample from one product in one category, but it is the kind of question worth adding to your own intake form before you decide how much this matters to you.
Does having an AI clone help you rank in AI search?+
Not directly, and we sell them, so take that as the honest answer rather than the modest one. There is no known mechanism by which having a clone causes an engine to cite you. What it does is produce the inputs: a continuous log of the real questions your audience asks, phrased their way, which is both the content brief and a source of original data that nobody else can publish.
Find out what your clients are actually asking
Build a clone of your expertise, put it in front of your audience, and get every question they ask logged and countable. That list is the content brief for everything above.
Free tier available. No credit card needed.