Your buyers stopped Googling. They ask ChatGPT, Gemini and Perplexity "what's the best tool for X", and the model names three brands. If you're not one of them, you never enter the conversation. No click, no visit, no chance to pitch.
Most founders don't know where they stand. They assume they're in the answer because they rank on Google. Those are two different games now.
This is the exact audit I run before I write a single page. It takes about 20 minutes, needs no software, and it works for any brand in any category. Copy it.
What you need
↳ ChatGPT, Gemini and Perplexity open in three tabs. Free accounts are fine.
↳ A blank doc with four columns: Question, Who got named, Sources cited, You in it? (yes/no).
↳ One competitor you lose to. You already know who it is.
That's the whole kit.
Step 1: Write the questions your buyer actually asks
Not keywords. Questions. The way a real person types into a chat box when they don't know your brand exists yet.
Aim for 10 to 12. Mix the intent:
↳ Recommendation: "What's the best [category] tool for [specific buyer]?"
↳ Alternatives: "What are the best alternatives to [competitor]?"
↳ Comparison: "[Competitor A] vs [Competitor B], which is better for [use case]?"
↳ Problem-first: "How do I [the job your product does] without [the pain]?"
The rule: your brand name never appears in the question. You want to see who the model picks when the buyer has no idea you exist. That is the honest test.
Step 2: Run every question through all three models
Paste each question into ChatGPT, Gemini and Perplexity. For each answer, fill your four columns:
↳ Which brands got named, in order.
↳ Which sources the model cited or linked (Perplexity shows these clearly, ChatGPT shows them when it searches, Gemini links some).
↳ Whether your brand made the list. Yes or no. No partial credit.
Run the same question twice if you have time. Models are not deterministic. A brand named on the first run can vanish on the second, and that instability is itself a signal that no one owns the answer yet.
Step 3: Score yourself honestly
Count it up. If you got named in 2 of 12 questions, your AI visibility is 17%. Write the real number down. This is your baseline, and it is the only number that matters until it moves.
Now look at the pattern, not the score:
↳ Which questions do you lose every single time?
↳ Which competitor shows up in the most answers?
↳ Are the questions you lose close to purchase, or just curiosity?
Losing a "what is [term]" question matters less than losing "best [category] tool for [your exact buyer]." Rank your losses by how close the buyer is to paying.
Step 4: Read the sources, because that's the whole game
This is the step everyone skips, and it's the one that pays.
AI answers are not opinions. The model runs a search, pulls a handful of pages, and writes its answer from them. So look at what it pulled. You'll see the same shapes again and again:
↳ A listicle: "11 best [category] tools in 2026." Your competitor is on it. You aren't.
↳ A comparison page pitting two rivals against each other. Neither is you.
↳ A Reddit thread where someone asked your exact question two years ago.
↳ A category definition page that quietly decides which brands count.
Those pages are the citation. The citation is the ranking. If you know which pages the model reads, you know exactly where you need to appear. You're not guessing at content anymore. You have a target list.
Step 5: Pick one gap and close it
Don't try to win everything. Take the single question that is closest to purchase and that you lose every time. Then give the model a reason to name you.
That usually means one page that does four things:
↳ Answers the question directly in the first 100 words, before any preamble.
↳ Includes a comparison table or a clean list the model can lift word for word.
↳ Names competitors honestly, including where they beat you. Models trust sources that admit trade-offs.
↳ Uses real numbers and specific claims, no vague superlatives.
The goal is not to rank on Google. The goal is to be the passage a model copies into its answer. Write for the lift, not the click.
Step 6: Re-run the same questions after you publish
Wait a week or two for the page to get indexed and crawled, then run your exact 12 questions again. Compare to your baseline.
↳ Before: named in 2 of 12.
↳ After: named in 5 of 12.
↳ Which source moved. Which competitor still owns the rest.
This before-and-after is the only honest way to report AI visibility. A dashboard that shows a score without a delta is a vanity metric. The delta is the proof.
Want the whole loop to run itself?
The audit above is real work. Twelve questions across three models, twice each, is 72 prompts, plus reading every source by hand. Doing it once is fine. Doing it every week for a moving target is a job.
That's what I built Fulcru for. There's a free Fulcru Skill you can drop into Claude, Codex or Cursor that runs this exact loop: it generates the buyer questions, runs them across ChatGPT, Gemini and Perplexity, records who got named, pulls the sources, and hands you the one page to publish next. Then it re-measures after you ship. The free visibility report at fulcru.app does the first pass for you, no card required.
One more thing worth flagging early. Step 4 keeps landing on the same truth: the brands AI names are the ones cited by pages the model trusts, and a lot of those citations come from third-party sites you don't own. Earning those placements by hand is slow. We're opening early access to a backlink and citation marketplace inside Fulcru that matches your pages with relevant, real sites the models already read. It's not live for everyone yet. If that's the part you care about, run the free report and you'll be first to hear when it opens.
Run the audit today. Even the manual version tells you something you didn't know this morning: whether the AI knows your name.
Want to see which questions ChatGPT, Gemini and Perplexity already answer with your competitor's name? Run the free Fulcru visibility report at fulcru.app.
Top comments (0)