Your buyers now ask an AI assistant for a shortlist before they ask a vendor. We count where four engines cite you today, run the work, and count again every month, so the number you take to the board is a measurement. You see movement or you see the plan change.
The usual pitch is structured data. No search vendor has stated that its AI reads FAQ markup, and the most-cited research on AI citations tested writing changes and never tested schema. The budget goes on the part that leaves the number where it was.
Your buyers are asking Perplexity, Gemini and Claude as well as ChatGPT, and the four disagree constantly about who to name. A dashboard on one engine can call you a leader or a laggard in the same week, and the board deck inherits whichever it picked.
Without a count taken first, no improvement can be demonstrated afterwards. Six months in, the question of what the money bought has no answer with a number in it, and the programme loses the room.
The Legency ApproachWe ask four engines the questions your buyers ask, repeat each one to filter out the noise, and count how often you are named and who gets named instead. That number is the contract.
ChatGPT, Perplexity, Gemini and Claude, asked the same prompts on the same cadence, so month four compares to month one. Where an engine is too unstable to claim movement, the report says so instead of banking the wobble as progress.
The questions your buyers type, answered in the first two sentences under the heading. Passages short enough to lift whole, a source behind every figure, and markup kept because it still earns rich results in ordinary search.
Four engines, the questions that matter to your market, each repeated for stability. It returns your share of answers, the competitors named ahead of you, and the questions where nobody is named yet, which is the cheapest place to start.
The pages that should be cited, rewritten to be liftable: the question as the heading, the answer directly under it, short self-contained passages, and a source behind every figure.
Schema per page type, an llms.txt that resolves, clean semantic structure and no broken internal links. It will not win a citation on its own, and its absence can cost you one.
The same panel re-run and reported beside rankings and AI-referred sessions. One document, in the same shape every month, written so a marketing leader can forward it to a CEO or a board as it is.
Every figure below is one client's August run: 14 buyer questions, each asked three times of four engines, 168 answers. The client is not named here; the numbers are exactly as the panel returned them. For a whole-category view, see the AI Visibility Index.
The same fourteen questions, asked of four engines in the same week, produced a 24-point spread. Perplexity cited this client in nearly half its answers. ChatGPT cited them in under a fifth.
That is the whole argument against a single-engine tool. A dashboard watching only ChatGPT would have reported this client as a laggard. A dashboard watching only Perplexity would have reported them as a leader. Both would have been describing the same fortnight.
It is also why the work is prioritised per engine rather than run as one undifferentiated content push.
PerplexityClaudeGeminiChatGPTA share figure on its own says nothing. The question a board asks is who is being named instead of us, and by how much.
In the August run this client was cited in 52 answers against AlphaSense's 50, having been behind in every previous run. That is a sentence with a number in it, which is the only kind worth putting in a report.
The same panel names every company an engine cited instead of you, so the competitive set is measured rather than assumed.
Ask an engine the same question three times and it will not always give the same answer. We measure that directly: stability is how often an engine repeated itself across its three runs.
Perplexity repeated itself every time. Gemini agreed with itself on fewer than three quarters of questions. So a Gemini figure that moves five points between months has not necessarily moved at all, and we say so in the report rather than banking it as progress.
Most tools in this category report the number and stop. The stability column is the difference between a measurement and a screenshot.
PerplexityChatGPTClaudeGeminiWe agree the fourteen questions your buyers ask, run them three times across all four engines, and hand back the baseline: your share, the companies named instead of you, the questions nobody owns, and the stability of each engine. Nothing is changed on the site this month. The number has to exist first or nothing afterwards can be proven.
The baseline names the questions where you lose. Those pages get rebuilt to be quotable: the question as the heading, the answer in the first two sentences under it, passages short enough to lift whole, and a source behind every figure. In parallel the technical layer gets fixed, because an engine that cannot cleanly separate your content from your navigation has nothing to quote.
The same fourteen questions, the same three runs, the same four engines. Movement against the baseline, per engine, with the unstable ones flagged rather than counted. This is where the engagement either shows a result or shows us that the plan needs changing, and both of those are reported.
From month four the cycle repeats monthly, and the report carries the trend rather than the snapshot.
A citation baseline is only as good as its question set. Ours is built from what buyers type, agreed with you before anything runs, and then frozen so month two is comparable to month one.
We never ask an engine about you by name. "Tell me about [your company]" is a question only someone who already knows you would type, and every engine will produce a flattering paragraph in reply. It makes a lovely screenshot and it measures nothing.
A tool whose report opens with your brand name in the prompt is measuring its own question. If you are evaluating anyone in this category, ask to see the prompt list before you sign.
Every question in our set could be asked by someone who has never heard of you. That is the entire point.
Both are worth money and they do different jobs. Buying one expecting the other is the commonest expensive mistake in this category right now. The same confusion sits between the disciplines themselves; our piece onAEO vs SEO separates the two.
A dashboard, a number, and a trend line. Cheap, fast to set up, and genuinely useful for watching. Most track one or two engines, most report mentions rather than citations, and none of them will rewrite a page.
Buy one if you have a team who will act on it.
The measurement, and then the work the measurement points at: the pages rebuilt, the technical layer fixed, the questions chosen, and a judgement each month about what to change. The number is the input to the job rather than the deliverable.
Buy this if nobody has time to own it.
We are not neutral here, so the honest test is simple: if you have a capable content team and a spare half-day a month, a tool plus your own people will get you a long way. If that half-day does not exist, a dashboard becomes another tab nobody opens.
No search vendor has ever stated that its AI reads FAQ markup, and the most-cited research on AI citations never tested schema at all. We keep structured data because it still earns rich results in ordinary search. That is the honest reason, and it is the one we give.
Engines re-crawl and re-index on their own schedule. First movement usually shows in the second or third monthly run. Anyone quoting you a number of weeks is describing something that is not theirs to promise.
Where an engine is too unstable across its own runs to support a claim, the report says so. A figure that moved inside the noise is not a result, and putting it in a deck is how a programme loses the room six months later.
If there are ten pages and no publishing cadence, the content has to exist before an engine can quote it. That is a different engagement and we will say so on the call rather than three months in.
ChannelSight is named in 34 of 168 AI answers across four engines, more than MikMak, Wayvia and Pear together. Every one of those numbers comes from our own monthly panel run.
A call about the questions your buyers ask, where you think you stand, and what a citation baseline would measure for your market.
It gets your company named in the answers AI engines give, and proves the change with a number a board can read. In practice that means measuring where you are cited today across the engines your buyers use, rewriting the pages that should be quoted so they can be, fixing the technical layer that stops an engine reading you, and re-measuring on a fixed cadence.
SEO competes for a position on a results page. AEO competes to be the source an engine quotes when it answers instead of listing. The underlying work overlaps heavily, which is why running them as separate vendors wastes money, but the target and the measurement are different.
Generative engine optimization and answer engine optimization describe the same work. GEO is the term the academic research uses, AEO is the term the industry settled on. Anyone selling them as two services is selling the same thing twice.
We ask each engine a fixed set of buyer questions, repeat each question so a one-off answer cannot skew the result, and record whether you were named, who else was, and what was cited. Repeated monthly on the same prompts, that produces a trend rather than a snapshot.
Engines re-crawl and re-index on their own schedule, so first movement usually appears in the second or third monthly run. Anyone promising results in weeks is describing something they cannot control.
Yes, and the evidence points that way. The most-cited research on AI citations tested content changes, not markup, and found that adding quotations, statistics and sources lifted citation rates the most. We keep structured data because it still earns rich results in ordinary search. We do not sell it as the AI play.
B2B companies with a real content estate and a marketing team to work with. If a site has ten pages and no publishing cadence, the honest answer is that the content has to exist before it can be cited, and that is a different engagement.