You asked ChatGPT the question your customers ask, and your product was not in the answer. Neither was it in Claude's, or Perplexity's. This page is for that exact moment. Two things up front, both measured: a zero is normal — most products in most categories are absent from most answers — and a zero is movable, because AI recommendations are built by reading a finite set of pages that can be changed. We measure this for a living and publish the method; here is what the fix actually consists of, and what it does not.
No tool inserts you into an answer — anyone selling that is selling something they cannot deliver, because the engines decide from what they read, not from who paid. What a tool can do is three real jobs: show you the answers your buyers actually get, show you the sources those answers were built from, and measure whether your changes move anything. The work in between — being present and well-described on the pages the engines read, and publishing the answers nobody else has written — is yours, and it is concrete. Start by measuring the zero properly: the free check runs your category's buyer questions across ChatGPT, Claude, Gemini, Perplexity and Grok and shows the verdict per engine.
Ask an AI about this page
Absent from all engines is a different problem from absent from some — and mentioned is a different thing from recommended.
Run the same buyer question across all five major engines, several times each. The engines disagree with each other constantly, so a single-engine check misleads: plenty of products are recommended by two engines and invisible to the other three. Which engines skip you, and on which phrasings, tells you where the work is.
And score what answers actually do, not whether your name occurs. In our corpus of 29,511 classified sightings, only 26.4% of the time a product was named was it explicitly recommended — the rest was listed among options, mentioned in passing, or occasionally warned against (the full split). “Mentioned” is not the goal. Recommended is the goal.
The engines built your category’s answer from specific pages. Those pages are the entire fix surface.
There is no submission form at OpenAI and no toggle at Anthropic. The answer changes when the pages it is built from change. Finding those pages is a measurable job — we published the full method — and the result is usually a short, workable list: the directories and review platforms the engines keep citing, the two or three editorial rankings that anchor the category, and the community threads that carry buyer trust.
Expect the list to be less independent than it looks: 48.8% of the citations in AI answers point at a vendor competing in that same category (90,960 of 186,331, 28-day window) — and in the one category we resolved end to end, all 22 of the most-cited pages were vendor-owned. That is not a reason for despair; it is the mechanism. Vendors are cited because their pages answer the buyer's question in plain language on their own domain. Yours can too, and for questions nobody has answered well, that is the fastest zero-mover we measure.
The cheapest source you control is your own site — if it answers the actual question, literally.
Take the buyer phrasings your category’s answers respond to, and check whether any page on your site answers them in those words. Not your feature page — a page whose heading is the question and whose first paragraph is the answer. Engines retrieve pages that match the asker’s language; founders routinely describe themselves in vocabulary their buyers never use, and become unretrievable to their own market.
This is also where honest differentiation pays. If your product is the one that works without an integration, or on-premise, or for a team of two, say it in exactly those words — the engines segment buyers hard, and specific fits win specific questions long before you win “best overall”.
Engines re-crawl and re-decide continuously. The fix is a loop, and the loop needs a baseline with a date on it.
Record today’s zero properly — which engines, which questions, which date — and re-run the same set on a schedule. Expect weeks, not days: the path runs through pages being published, crawled and absorbed into answers. What you are watching for first is not the recommendation; it is your sources appearing in the citation lists. Citations lead, mentions follow, recommendations come last.
And keep the goalpost honest as you climb: appearing in answers is progress, being recommended is the win, and being recommended for the right buyer is the durable version — the engines actively steer people away from products that are wrong for them, 614 times in our last 28-day window alone (the Warning List).
No. There is no placement product for organic AI answers, and any vendor promising guaranteed inclusion is promising something outside their control. What you can buy legitimately is measurement, the work of getting present on the sources engines read, and content that answers buyer questions — all of which move answers indirectly, none of which is a paid slot.
When it moves, it moves on crawl-and-refresh timescales — typically weeks. Anything promising days is describing paid placements on citation sites, not changed answers. Measure weekly and judge on the trend, not any single run.
Sometimes, and less often than assumed — it depends entirely on whether those platforms are in YOUR category’s citation list. In categories we have resolved, the most-cited pages are frequently vendor sites and editorial rankings, not the big review platforms. Check the reading list first; join second.
Yes. The free check runs your category’s buyer questions across all five engines and shows the verdict and sources per answer, no card required. That is the baseline everything else is measured against.
Numbers on this page: mention-vs-recommendation split from the measurement-problem study (29,511 sightings, 4,162 products, 105 categories) · vendor-owned citation share 48.8% (90,960 of 186,331, 28-day window, pulled 2026-08-17) · top-cited-pages finding from Category Influence Audit #1 v1.2 (125 answers, 1,188 resolved citations) · warning counts from the Warning List study (614 warnings in 189,375 appearances, 28 days ending 2026-08-26). Live values on /data.
You cannot fix a zero you have not measured. Start with the real answer set.
Measure Your Zero Free →© 2026 Orbator. All rights reserved.