Rankings still hold.
The clicks left anyway.
Measure how often assistants name your brand, read every page those answers cited, and get the gaps back as work you can brief. It adds the surface your stack cannot see — it does not replace it.
Want to see the method first? See AI sources
- Answers stored in full, not just scored
- Every cited page kept per answer
- Free audit, no card
Rank held, the click did not, and nothing counts the answer
Being named inside an answer is the new placement. Your instruments still only count positions and links.
-
Position one, and still not a single visit
The overview answers in place. Impressions hold, clicks fall, and rank still looks fine.
-
Being read is not being named
On one run the assistants opened 16 pages of the brand's own site and 8 of those answers went on to name somebody else. The authority was there and the format was not — which is an edit, not a new page.
-
Mentions are counted nowhere in your stack
A brand named inside an answer is a placement, and no tool you own records it. There is no impressions column for it and no Search Console row.
-
Your brief has no evidence sitting behind it
"Write for AI" is advice, not a spec. The team needs the answer, the claim missing from it, and the page to change.
The same rigour, on the new surface
Mentions become a tracked metric
Named or not, in what position, on which assistant, with the sample printed beside it and history behind it.
Every answer keeps its citations
The pages cited alongside each answer, sorted by who owns them. About half are ones you could publish on; a rival's own site never is, so it is marked instead of counted as an opportunity.
Gaps become briefs
A missing claim is a rewrite. A listicle naming a rival is an outreach target. Each one arrives with the answer that produced it.
It fits the process you already have
Four steps from a domain to the citations sitting behind who gets named.
-
The questions get found, not mapped from your keywords
A keyword and a buyer question are different objects, and mapping one onto the other quietly changes what you measure. These come from Reddit threads, People Also Ask, search volume and the rivals already ranking. Sixty are tested, five are kept, and every rejection is shown with its reason.
A question set you can argue with, line by line
Step 01
-
Run them across every engine
Five questions, four assistants, three runs each. Sixty answers a reading, every one stored with its engine, run number and citations.
Coverage across four assistants, on a schedule
Step 02
-
Read the citations back
Which domains keep getting opened in your category, which are yours, and which you could realistically get onto. On one run 47% of the source pool was a rival's own site.
The domains the assistants keep returning to
Step 03
-
Brief the fix to your team
Each gap comes out as one move, cheapest first, naming the page or the thread. The same locked questions run again weekly so the next reading measures the same thing.
A ticket-ready fix with its evidence attached
Step 04
The parts that replace the missing instrument
-
Competitor Tracking
The brands winning the answer on your own question set, read out of the answers rather than a list you typed, tracked week to week like any other placement.
-
Prompt Discovery
Locked buyer questions asked three times each, every raw answer and citation kept, and a single lucky mention re-run nine more times before it counts.
-
AI Visibility
Mention rate, position and share of voice per assistant, never blended, with the number of answers printed beside every figure.
-
AI Sources
Which pages get cited, how often, and who owns them — the closest thing to a backlink report for answers. It is where the assistant was reading, not why it picked anyone.
-
Growth Opportunities
Gaps written out as briefs, each naming the page, the claim to add, or the source to go and earn.
SEO team questions
The ones that come up in the first evaluation call.
Does this replace our rank tracker? +
No, and you should be sceptical of anything that says it does. Rankings still decide the classic result and you should keep tracking them. This measures the layer above — whether the answer names you at all — which no rank tracker looks at.
Google says good SEO is good GEO. Is that not enough? +
It is a floor, not a ceiling, and our own runs show where it stops. On one run the assistants opened 16 of the brand's own pages and 8 of those answers still named a competitor. On another, the domain's site was cited in 9 of the 32 answers that named the top rival. Ranking got the page retrieved; it did not get the brand recommended.
How do you know which sources an answer used? +
We keep the citations the assistant returns with the answer and normalise them to domains and pages. Where an assistant returns none, we say so rather than inferring one. What we will not do is tell you which sentence came from which page — the assistants do not report that.
Can we bring our own prompt set? +
There is no bulk import today. Questions are found for you and shown with their source, and any one of them can be struck out — the correction is stored against your domain, so it survives every future run instead of being redone each time.
Can you use our Search Console data? +
Yes, as an optional connection, and it pays out immediately without a re-run. Your ranking queries are split from your brand queries, grouped by distance — won within position three, close within ten, reachable within twenty — and each is judged against what you actually sell, so keywords that rank but are not your business stop inflating the plan.
Is the sampling consistent enough to trust a trend? +
Same locked questions, same schedule, same assistants each run, three runs per question, and every raw answer stored. If a number moves you can open the answers from both weeks and read what changed. Worth knowing that two runs of one question return the same brand list less than once in a hundred, which is exactly why one reading is never reported as a trend.
Does it cover AI Overviews as well as chat assistants? +
Yes. Google's AI Overviews is tracked alongside ChatGPT, Perplexity and Gemini, and scored on its own because it behaves differently. On one run a brand sat at 4 of 13 on AI Overviews and 0 of 15 on ChatGPT, same questions, same day.
Can we export the data? +
Not as a bulk file yet. Every answer is stored in full and opens on a link you can send to anyone, and the report is available over the API. A file export is not shipped.
Put a number on the traffic you cannot see
Run the free audit on your domain. You get the mention rate per assistant, the rivals taking the slot, and every page those answers cited.