Know which AI answers name your business, and which to win next.
Ask ChatGPT, Gemini, Claude and Grok the questions your buyers type, and read back which answers named you. The score, the split by engine and every question sit on one screen, with the date of the run beside them.
How many AI answers name my business?
The score is the share of answers that named you, out of the answers received. Your team writes a set of ten to fifteen questions during setup, worded the way a buyer types them. Each run puts every question to ChatGPT, Gemini, Claude and Grok through its own API. Twelve questions and four engines make 48 answers. Get named in 24 of them and the card reads 50 / 100.
Open the card and read the run behind it. The note states the date, the number of questions asked and the exact count of answers that named you. The change since the last run sits beside the score, and a small trend line carries the recent runs. The first run reads baseline. A later run with no movement says no change rather than inventing one.
Whether an answer named you is decided from the answer text. The check looks for your business name in a form written to avoid a common word or surname, so a stray match is not counted as a citation. The first 1,200 characters of every answer stay in the run record, so your team can audit any verdict and read who the engine named instead.
Can I trust the number?
Yes, because a blank answer never counts as a miss. An engine that returns nothing has not declined to name you; it has said nothing. Scoring that blank as not cited would manufacture a confident zero. So each blank is recorded as a measurement error and leaves the denominator.
Take one question put to four engines. ChatGPT names you, Gemini does not, Claude names you, Grok returns nothing. That question scores 2 of 3, not 2 of 4. The run record keeps the error text for every excluded answer, and the card's note states how many answers were actually received, so the denominator is on the screen rather than in a footnote.
The change figure is guarded the same way. When the measurement method changed, with web search turned on for every engine, the next run read baseline instead of a delta, because a difference the work did not cause is not progress. Ask any vendor what their score does with a null response. A score that cannot answer is not a measurement.
Which questions do I win, and which do I lose?
Every question is listed, won questions first. Each row shows the question as it was asked, a cited badge where an engine named you, and the count of engines that did out of the engines that answered. Click a row and it opens: named or not yet, and which engines were asked. The won list says what is working. The not-yet list is the work plan.
Expect a spread. On one measured account, all four engines named the business for one phrasing about its trade in its own town, and none of them did for another phrasing about the same trade in the same town. A set skewed to one phrasing measures the phrasing, not the business. So your set spans the ways a buyer asks: the trade, the problem, the urgency, the towns you serve, and the commercial work if you do it.
Change the set through your team. Ask from the screen, through the content or keyword request, or by email. The request is logged and the team edits your set. The set stays fixed between runs, so when the number moves, something changed on the web and not in the questions. Nothing on the screen writes to the set, your listing, your site or your ads account.
Which engines are asked, and how?
ChatGPT, Gemini, Claude and Grok, each through its own API, each given the same question. Web search is on for every one of them. An ungrounded model returns what it memorised in training, and that is not what a buyer sees. People ask in the apps, and the apps search the live web and read your Business Profile, the directories and the review sites. A business whose strength is local presence and reviews is exactly the business an ungrounded score would understate.
The question goes in bare, the way a person types it. No preamble asks the engine to answer as a homeowner or a procurement manager. A wrapper like that changes what the engine retrieves, and the run would measure our prompt instead of your buyer's experience.
At every run the ChatGPT, Gemini and Grok model versions are checked against each provider's live catalogue, so a retired version cannot keep scoring in silence. For Gemini the rolling alias is used on purpose, because Google moves that target. The run record names the model that answered. Google's AI Overviews are a search result, not an answer-engine answer, and this view does not score them.
Presence by engine sits beside the score. Strong in one engine and absent in another is common, and it is a different fix from being absent everywhere.
What moves the number, and who does the work?
The same things that move search, applied to how an engine reads a business. Facts about you that agree everywhere the engines look: your Business Profile, the directories and the review sites. Schema markup on your service pages that says what you do and where. A claimed and complete Business Profile with steady reviews, because engines lean on review signals for local trades. Pages with answer-shaped headings and FAQ schema, written to be quoted. That work is the program, described on the AI search services page. This view is the instrument that says whether it worked.
Part of it runs on a schedule whether anyone is at a desk. Where the content engine runs for you, every piece it writes opens with a direct answer of 40 to 60 words built to be lifted and quoted, names your business and the area you serve early in plain words, closes with a question-and-answer section, and files the draft as a Google Doc in your Drive. The not-yet list tells your team which questions those pieces should answer. The rest stays with a person: writing your question set, reading who was named instead, and changing what your site and your listing say.
The baseline is taken in the second week of setup, alongside your keyword set. Setup takes two to four weeks and is not billed. The set is re-run as the program moves, and the card carries the date of the run it shows. The AI visibility program is on all tiers.
What should I check in the demo?
Open demo.trulata.com. There is no form. The company is fictional and every number is illustrative, but the engine code is the same code every client instance runs. Open the AEO / GEO view. Click the score card: the detail opens with the score, the change and the top model. Click a question row: it opens with named or not yet and the engines asked. On your instance every number on the view comes from a run record, and a short note under the question table says what your set spans and which work moves it.
Then hold it to the honest ceiling. The view reports named or not named; it does not report a position inside an answer. The Refresh button on the dashboard re-reads the last scoring run; it does not put the questions to the engines again, and the card's date says when that last happened. No question set is scored on a clock. The card carries the date of the run it shows. Read the change figure against the dated run beside it, never on its own.
Nothing needs connecting for the measurement itself. The engines read the public web the way a buyer's app does. The program side uses your Business Profile and your site. The Business Profile connects when you add our Google account as a manager on the listing, an invitation you can revoke in your own console. We never hold your passwords.
What the AI visibility view reads.
| What | Where it comes from | How fresh |
|---|---|---|
| Questions | Ten to fifteen, written with your team in setup week two | Fixed between runs; changes through your team, logged |
| Answers | ChatGPT, Gemini, Claude and Grok, own APIs, web search on | As of the run dated on the card |
| Named or not | Your business name found in the answer text | Per run; the first 1,200 characters kept |
| Score | Answers naming you over answers received; blanks excluded | Per run, with change or baseline |
| Presence by engine | The same run, split by engine | Per run |
A blank answer is a measurement error, never a miss: it leaves the denominator, and the card states how many answers were actually received.
Questions, answered.
Which AI engines do you measure?
ChatGPT, Gemini, Claude and Grok, each through its own API with web search on and the same question set. Google AI Overviews are a search result and are not scored here.
What is the score out of?
The answers actually received. A blank answer is recorded as a measurement error and left out of the denominator, and the card's note states the count of answers received.
Can I see the questions?
Yes, every one, won questions first, with the count of engines that named you and a cited badge where one did.
How often is it re-scored?
At baseline in setup week two and again as the program moves. The card shows the date of the run, and the Refresh button does not re-run the engines.
Can I add questions?
Yes, through your team. Ask from the screen or by email; the request is logged, the team edits the set, and the set stays fixed between runs so the number stays comparable.
Is this the same as SEO?
No. This view measures whether an answer engine named you. The search view measures average positions, queries and clicks. Both sit on one screen because a buyer uses both.
See it running
before you decide.
The demo is the real product on a fictional company, with no form in front of it. Pricing is three published tiers. Setup on your own accounts takes two to four weeks.
Open the live demo See pricing
New here? See what the Command Center is. Prefer to write? Send us the one thing you want to know.
- Refreshed when you open itLive data on page load, never a monthly PDF
- Counted from your own formsLeads recorded server-side, compared daily with Ads and GA4
- Four answer engines, measuredChatGPT, Gemini, Claude and Grok, with a published denominator
- A review-first queue on every sendEvery draft waits in a review-first queue and every send is logged.