How long does the manual check take?

20 minutes if you follow the recipe. 10 of that is pasting prompts into four AI apps; 5 is writing down what you see; 5 is interpreting the result. Faster the second time you run it because the prompts and the scoring sheet are already set up.

Do I need accounts for all four AI engines?

A free account on each. ChatGPT and Gemini have free tiers. Claude has a free tier. Perplexity works without an account but you'll get better answers signed in. None of this costs anything. The whole exercise is free.

What if the AI gives a different answer when I run the same prompt twice?

Normal. AI responses are non-deterministic, so the same prompt can return different answers minutes apart. That's why this manual check is for a baseline reading, not for trend data. If you need to compare results across time, automated monitoring is the only way that holds up.

Can I do this for my competitors too?

Yes, and you should. Run the exact same prompt set you used for yourself, but compare how each competitor appears in the same answers. The competitor data is usually more revealing than your own, because it shows you which brands the AI considers default in your category.

What's the difference between this and using a free AI visibility checker?

A free checker automates the entire recipe in 30 seconds. The manual version takes 20 minutes but teaches you how the engines actually behave (which is useful context the first time). Most people run the manual version once, then default to the automated checker for everything after.

How often should I rerun this manual check?

Once for the baseline, then only when something specific triggers a recheck (you published a major piece of content, a competitor launched, your category got a wave of press). For ongoing tracking, the manual method doesn't scale; that's what monitoring tools are for.

All Guides

Check your AI search visibility in 20 minutes

A step-by-step recipe. No budget, no tool, just a browser, four free AI accounts, and a scoring sheet you can copy from this page. The version of the work that fits in a coffee break.

Matiss Katanenko

Co-founder, Honeyb

Last updated May 16, 2026

What this recipe is for

This is the version of the work for someone who needs to know, today, whether AI is recommending their brand or their competitors. No budget approval, no procurement, no tooling. Just a browser and 20 minutes.

It produces a useful baseline reading. It will not produce trend data, and it isn't reportable to a board, but those aren't the goals at the start. The goal at the start is to find out whether AI visibility is your problem or somebody else's, and what size it is.

If you'd rather skip the manual setup and have this done for you in 30 seconds, the free AI visibility checker runs exactly this recipe across all four engines automatically. The manual version below is for people who want to see how the engines behave before they pick a tool.

What you need

A web browser
A free account on ChatGPT (chat.openai.com), Gemini (gemini.google.com), Claude (claude.ai), and Perplexity (perplexity.ai). Perplexity works without one but signed in gives better results.
A spreadsheet or notepad. There's a copy-able scoring sheet later in this guide.
20 minutes.

The five steps

Each step has a target time. If you find yourself overrunning step 1 by more than five minutes, the prompts are getting overcomplicated. Stop and pick simpler ones.

Pick three prompts

5 minutes

Write one category query, one comparison query, one use-case query. The whole exercise depends on the prompts being the questions your buyers actually ask.

Category query: "best [your category] for [your buyer type]". Examples: "best CRM for SaaS startups", "best e-commerce platform for a DTC brand".
Comparison query: name two or three competitors you genuinely compete with. "Pipedrive vs HubSpot for small sales teams", "Shopify vs BigCommerce vs WooCommerce for fashion brands".
Use-case query: a problem statement. "What's the right tool for a 5-person agency tracking client retainers", "what should I use to do email marketing if I'm just starting".
Use the language buyers use, not the language you use to describe yourself. "AI-native customer engagement platform" is not a query anyone types.

Run each prompt on all four engines

10 minutes

Paste each prompt into ChatGPT, Gemini, Claude, and Perplexity in turn. 12 total runs. Don't refine the prompt mid-run, don't tell the AI it got the answer wrong, don't ask follow-ups. The first response is what a real buyer would see.

Open each engine in a separate tab. Working through them in parallel rather than one at a time is faster.
Copy the exact same prompt into each. Paraphrasing skews the comparison.
If an engine asks a clarifying question, give the most generic reasonable answer ("a small business", "a B2B SaaS company", whatever's relevant) and proceed.
Record the response. Screenshot or copy-paste the text into your scoring sheet.

Record what you got

5 minutes

For each of the 12 cells (3 prompts × 4 engines), note four things. Don't analyse yet; just capture.

Did your brand appear? Yes / no.
If yes, in what position? First mentioned / middle of the list / last / passing reference.
With what framing? "Recommended", "one option among several", "mentioned with a caveat" (e.g. "X is good but pricing is high"), or "named but not endorsed".
Which competitors appeared in the same answer? Note the full list.

Interpret the result

Optional, 5 minutes if you have time

Now you can look at patterns. Three questions to ask.

What's your mention rate per engine? (You appeared in 2/3 ChatGPT answers = 67% rate.)
Where's the cross-engine variance? If you're on three engines and missing on one, that's a targeted problem. If you're missing on three of four, the gap is general.
Who are your real AI-visibility competitors? Often different from your stated competitor set. The brands that appear in the answer set are the ones the AI considers default in your category.

Decide your next move

Decision, not time-boxed

Three honest reactions depending on what you found. Pick one and act on it. Don't try to do all three.

Invisible everywhere: third-party validation is the biggest lever (review platforms, editorial roundups, community presence). Don't reach for content production first.
Visible but framed wrong: structured data on your own site, plus finding which third-party source the AI is pulling outdated information from.
Visible and framed well: set up monitoring on those exact prompts so you'll notice when the pattern shifts. It will shift.

The scoring sheet (copy this)

Paste this into a spreadsheet. One row per engine per prompt. Twelve rows total for three prompts.

Prompt	Engine	Mentioned?	Position	Framing	Competitors named
Category query	ChatGPT	.	.	.	.
Category query	Gemini	.	.	.	.
Category query	Claude	.	.	.	.
Category query	Perplexity	.	.	.	.
Comparison query	ChatGPT	.	.	.	.
Comparison query	Gemini	.	.	.	.
Comparison query	Claude	.	.	.	.
Comparison query	Perplexity	.	.	.	.
Use-case query	ChatGPT	.	.	.	.
Use-case query	Gemini	.	.	.	.
Use-case query	Claude	.	.	.	.
Use-case query	Perplexity	.	.	.	.

How to read the result

Rough benchmarks for interpreting your scoring sheet. Not laws, but useful orientation.

0/12 cells with your brand: AI doesn't have you in working memory for this category. Common for newer brands, niche players, and brands whose third-party footprint is minimal. The fix is upstream, not on your own site.

1-3/12 cells: you exist in the model but you're not default. Often this means you're known to one or two specific source domains (a single review platform, a niche publication) and that source isn't getting cited consistently across queries.

4-7/12 cells: you're in the consideration set. Strong position, real room to grow. The work is converting the cells where you appear with weak framing into the cells where you're the recommendation.

8-12/12 cells: you're a default answer in your category. This is the win condition. The work shifts from \"earn share\" to \"defend share\" and to detecting when competitors start eating it.

The single most common mistake at this stage is rerunning the check three times until you get a result you like. Run it once, write down what you got, act on that. The check is a baseline reading, not a survey.

When to graduate from manual to monitoring

The manual recipe is for the baseline. Three signals it's time to set up a real monitoring loop.

You've run the same recipe more than twice. At that point, you're doing manually what a tool does in 30 seconds with better consistency.

Someone asked for a trend line. The first time a stakeholder says "how has this changed," the recipe can't help you. You can't reconstruct trend from a single baseline.

You want to attribute a content or PR change to a visibility shift. Attribution requires before-and-after on consistent measurement. The manual recipe is too noisy for that comparison to hold up.

Frequently asked questions

About the author

Matiss Katanenko

Co-founder, Honeyb

My name is Matiss Katanenko and I co-founded Honeyb, the AI visibility platform that tracks how ChatGPT, Gemini, Claude, Perplexity and the other major AI engines talk about brands. I'm based in Riga, Latvia. Before Honeyb I spent years on the agency side running SEO and content programs for fast-growing brands across the US and Europe. That work is where I watched AI search start to compress the entire discovery channel into a four-brand short list, and decided to build the tool I wished agencies had. In my free time I'm in the sauna, on a padel court, or behind a drum kit.

Connect on LinkedIn

Free, instant, no signup