How to Check If AI Recommends Your Brand

You check whether AI recommends your brand by building a fixed set of 15 to 30 questions your buyers actually type, running them on a repeating schedule across ChatGPT, Gemini and Google AI, and scoring three things each time: whether you are named, where you sit in the answer, and which competitors appear beside you. The number that matters is not "are we mentioned," it is your citation share against a named competitor set, tracked over weeks, because a single spot check is noise.

That is the whole method. The rest of this piece is how to run it properly, why most brands fail the check, and how to turn the score into a pipeline number instead of a slide.

Start with the questions your buyers type, not your product name

Most visibility audits fail at the first step because they probe the wrong thing. Searching your own brand name in ChatGPT tells you almost nothing. Of course it knows you exist. The question is whether you get named when someone describes a problem and asks who solves it.

Build the prompt set across four framings:

• Category. "What are the best tools for X?"
• Comparison. "A vs B for a 40 person team."
• Use case. "How do I run paid search reporting that ties back to closed revenue?"
• Problem. "My leads, deals and campaigns are in five systems and nothing reconciles. What do I do?"

Write them the way a buyer writes them, with their vocabulary, their constraints, their company size. Fifteen is the floor, thirty is better. Then freeze the list. The set only works as a measurement instrument if it stops changing.

One honesty note before you over-trust the output. The most detailed public tracker I have seen, Bill Widmer's tracker of 13,184 citations across four models, published on Orbit Media with data through Aug 23 2026, used 12 to 40 questions per brand and the author is explicit that this is directional, not statistical proof. He also discloses that Semrush, one of the three tracked brands, is his client. Britney Muller, quoted in Orbit Media's off-site AI search optimization piece and referenced in that study, describes ten runs of a prompt as "a very crude directional signal." Treat your scorecard as a trend line, not a measurement to three decimal places.

Score three things per prompt, not one

Presence is a weak metric. Score each run on:

1. Named or not. Binary.
2. Position. First recommendation, in the list, or a passing mention at the bottom. Being third in a list of eight is a different commercial outcome from being the opening answer.
3. Who else is in the room. Log every competitor named alongside you.

That third column is where the budget argument lives. Across your full prompt set, count how many answers name you versus how many name each competitor. That is citation share, and it has a denominator, which means a CFO can read it. "We appear in 6 of 22 buyer questions, our closest competitor appears in 17" is a sentence that moves money. "Our AI visibility score is 42" is not.

Add two more columns while you are there. Thrive Agency's manual tracker format is the sensible baseline: date, prompt, answer, mentioned brands, citations, sentiment, inaccuracies. The inaccuracy column earns its place fast. Being named with outdated services or the wrong positioning is not a win, it is a liability you now know about.

Mention and citation are different metrics, and your size decides which one matters

This is the trap in almost every visibility dashboard on the market. Being mentioned in an answer and being cited with a link are not the same event, and the gap between them is large.

In Widmer's data, Claude mentioned the tracked brands in 70% of answers but cited them in 40%. Gemini: 72% mentioned, 41% cited. Perplexity: 67% and 44%. ChatGPT was the outlier at 56% mentioned and 47% cited, meaning it is the engine most likely to link what it names, though it also had the lowest mention rate of the four.

Brand size may distort this further. In the same study, Semrush was mentioned in 83% to 100% of relevant answers while its citation rate on those same answers ranged from 83% all the way down to 0%, roughly a 50 point mention-to-citation gap. The two smaller brands tracked showed gaps of 4 to 5 points. The author's read is that if you are a large known brand, citation-only dashboards will understate you badly, though he offers a competing explanation: Semrush has far more content than the other two sites, so the gap could be that the smaller brands were only mentioned when their name was in the prompt.

Score both columns. Decide which one is your headline number based on where you actually sit.

Run a panel, not a spot check

Answers move. The same 72 questions re-run two days apart kept only 30% of previously cited URLs in Claude and 38% of domains in Gemini. Perplexity was the most stable at 65%. Two days. Not two quarters. One caveat on the Claude figure: the author's Claude data comes from an agent-with-search setup rather than the raw API, so citation style may differ from claude.ai.

So a one-off audit screenshot, the kind that arrives in a cold email with your logo and a red X next to it, is weather. Four weeks of the same frozen prompt set is climate. Run weekly or fortnightly, same questions, same models, log everything, and read the trend.

Run it across multiple engines too, because they disagree. Across 1,792 query-and-domain combinations in that dataset, all four models cited the same domain for the same question only 30 times, 1.7%. They also differ in how much they pull in: Perplexity averaged 19.2 sources per answer, Gemini 8.1, ChatGPT 4.5, Claude 3.6. Fewer sources means a harder door.

And stop using your Google rank as a proxy. Domain-level overlap between AI citations and Google's top 10 for the matching query ran 10% to 30%. At exact-URL level it was roughly 1% to 5% for Gemini and ChatGPT. Perplexity tracks Google closest and still overlaps only 13% to 23% at the URL level. Page one is neither necessary nor sufficient.

Worth saying plainly: the instinct to go optimize per engine is mostly wasted effort at this stage. Nick Lafferty of Profound, quoted in the same research, put it as "we're not doing things differently for different engines."

Why AI skips you: adjective pages lose to checkable facts

Here is the part most guides get wrong. The reason you are not named is usually not keyword coverage. It is that your site does not state anything an AI can verify and repeat.

ChatGPT's sourcing in Widmer's data skewed toward primary sources like Google documentation, arXiv papers, and vendors' pricing and help pages. That is the tell. Pages that state specific, checkable things get retrieved. Pages full of "industry-leading," "end to end" and "trusted by growth teams" get skipped, because there is nothing in them to cite.

What to publish, in order of return:

• Pricing, in numbers, on a page. Not "contact us for pricing."
• Integrations, named. A list, not a logo wall.
• Process and methodology. What actually happens, in steps, with timeframes.
• Named use cases. The specific job, the specific buyer, the specific outcome.
• Documentation and help content. The sourcing data shows ChatGPT reaching for vendor help pages.

One more correction while we are here. The common advice to go build presence on Reddit does not hold for B2B buyer questions. In the last run of that study, Reddit accounted for 0% of Claude's citations, 0% of ChatGPT's, 3% of Perplexity's and 1% of Gemini's. Across hundreds of answers, Claude never cited Reddit once.

Payani's AI Search Visibility scoring exists for this layer specifically: it checks whether you show up, handles the technical items that affect how AI reads your site, and the article engine publishes the fact-dense pages that earn the citation. The work is retrieval-shaped, not keyword-shaped.

Make it a pipeline number

A visibility score that never touches revenue is a vanity metric with better branding. Close the loop:

1. Add a field on your forms asking how they heard about you, and tag inbound contacts who mention an AI assistant.
2. Watch whether prompt-set gains show up as sessions and form submissions, not just as a higher score.
3. Track those contacts through to deals, so you learn the win rate and deal size of AI-sourced pipeline, not just its volume.

In Payani, that chain is already connected: website visitor and company activity tracking, forms, lead scoring and the CRM sit on the same data, so a rise in citation share and a rise in qualified pipeline are visible in the same place rather than in two tools that never reconcile.

Start this week. Write 20 questions, run them once, log the three columns, and put a repeat in the calendar. In a month you will have a trend line, and the fix list will be obvious from the gaps.

Frequently asked questions

How many prompts do I need to measure AI visibility properly?

Fifteen is the floor and thirty is better, spread across category, comparison, use case and problem framings. Be honest about the limits: the independent Orbit Media study used 12 to 40 questions per brand and its author states clearly that this is directional, not statistical proof. Treat the output as a trend line, not a precise measurement.

Is being mentioned the same as being cited?

No, and the gap is large. In the Orbit Media data Claude mentioned tracked brands in 70% of relevant answers but cited them in 40%, Gemini 72% versus 41%, Perplexity 67% versus 44%, and ChatGPT 56% versus 47%. Large brands see much bigger gaps, so citation-only dashboards understate them, while small brands track roughly one-to-one.

How often should I re-run my prompt set?

Weekly or fortnightly, with the same frozen question list. The same 72 questions re-run just two days apart kept only 30% of previously cited URLs in Claude and 38% of domains in Gemini, with Perplexity most stable at 65%. One run is weather, four weeks of runs is climate.

If I rank on page one of Google, will AI name me?

Not reliably. Overlap between AI citations and Google's top 10 for the matching query ran 10% to 30% at domain level and roughly 1% to 5% at exact-URL level for Gemini and ChatGPT. Page one rank is neither necessary nor sufficient, so measure AI answers directly.

What should I fix first if AI never names my company?

Publish checkable facts rather than marketing adjectives. ChatGPT's sourcing skews toward primary sources including vendor pricing and help pages, so stated pricing, named integrations, documented process, named use cases and real documentation get retrieved while claim-heavy pages get skipped.