Back

Guide

How to Check Your AI Visibility for Free (and How to Increase It)

November 19, 20269 min read

Short answer: to check your AI visibility for free, write 10–20 questions your buyers really ask, put each one to ChatGPT, Claude, Gemini and Perplexity with web search on, run each more than once, and log whether you're named, who's named instead and which sources are cited, then repeat it monthly with the same questions. You can't edit the answers, but you can change what they're built from: your own pages, the lists and directories that rank, review sites and community threads.

"How do I get ChatGPT to recommend my SaaS?" comes up again and again in founder communities, and most of the answers jump straight to tactics. Measure first. Until you know which questions you're missing from, and who gets named instead, every fix is a guess.

What is AI visibility?

For a SaaS, it means three different things, and a useful check looks at all three:

  • Named: the assistant mentions your product when a buyer asks a question you should be the answer to.
  • Cited: one of your pages is among the sources the answer links to.
  • Described correctly: what the assistant says about you is current and is actually about you.

Most checks count only the first. The third is the one that surprised us most when we checked ourselves (more on that below).

How do I check my AI visibility for free?

1. Write 10–20 real buyer questions

Take them from what buyers actually say: sales calls, support tickets, the "what were you looking for?" answers in your onboarding, Reddit threads, the "People also ask" box in Google. Word them the way a buyer would, not the way your homepage would. Cover four kinds. For a made-up shift-scheduling tool for small clinics called Rotaline:

  • Category: "best shift scheduling app for a small dental clinic".
  • Problem: "how do I stop staff swapping shifts over WhatsApp".
  • Comparison: "[the tool you replace] alternative for small clinics".
  • Branded: "what is Rotaline", "is Rotaline any good".

Branded questions belong in the set even though they feel vain. They are where you find out whether you're described correctly.

2. Ask four assistants, with web search on

Use ChatGPT, Claude, Gemini and Perplexity. They read different sources, so being named by one says little about the others. Turn web search on wherever it's a setting (at the time of writing Perplexity searches by default, and the others search on many questions or when you ask them to). Search is the version that reads current pages, and the only one you can influence this month. Without it, a model answers from training data that is months old, and a young product usually isn't in it.

Start each question in a fresh chat, logged out or with memory and personalisation turned off where the assistant allows it. Your own history with an assistant can steer the answer toward the product you keep asking about, which is yours.

3. Run each question more than once

Answers vary from run to run, and with small changes in wording. One run is an anecdote. Run each question at least twice, three times for the category questions you care about most, and record "named in 2 of 3 runs" rather than yes or no. If you have time, add one rewording of each important question.

4. Log what you see

For every run, write down whether you were named, where and how, who was named instead, and which sources were cited. The sources column is the most valuable one, because it tells you which pages the answer was built from. That is your to-do list.

5. Repeat monthly, with the same questions

Keep the core set fixed and the settings the same, so this month compares with last month. Add new questions in a separate section as you learn how buyers phrase things. A check you run once tells you where you stand; a check you repeat tells you whether anything you did worked.

What should the log look like?

One row per question, per assistant, per month. Here's what a few rows might look like for the fictional Rotaline:

AssistantQuestionNamedHowNamed insteadSources cited
ChatGPT (search on)best shift scheduling app for a small dental clinic1 of 24th of 5, "simple, for small teams"Two large scheduling suites, one free toolTwo "best scheduling apps" lists, one review site
Perplexitybest shift scheduling app for a small dental clinic0 of 2—The same two suitesA review site, a forum thread
Geminiwhat is Rotaline2 of 2Correct, but quotes last year's price model—An old directory listing

Even three rows say something: two "best scheduling apps" lists and a review site are doing the work in the category answer, and an old directory listing is behind the outdated description. Neither fix is on Rotaline's own site.

What we found when we checked Ordana

In November 2026 we ran 15 buyer questions we would want Ordana to be part of, such as "where should I submit my SaaS after Product Hunt", "how do two SaaS companies split revenue from an integration" and "what is a widget integration", through a web search index. That is a proxy for what search-grounded assistants read, not the assistants themselves, so read it as "what the sources looked like", not "what ChatGPT said". What came back:

  • Ordana appeared in 0 of 15. No page of ours, and no snippet that mentioned us.
  • Even searching our own name missed our site. What came back were old third-party listings describing an earlier version of Ordana, from before we rebuilt it around widgets, and other companies with similar names. In one summary, the search tool blended us with one of them.
  • The index couldn't see Reddit at all. An assistant grounded on it would never read a Reddit thread, however good. Different assistants really do read different sources.
  • Five of the fifteen questions had no clear answer. The results were vendor documentation, forum pages or pages answering a different question. Those are the cheapest questions to win, because the competition is weak.

So the first fixes on our list weren't new content. One was our structured data: Organization sameAs links from our site to our own profiles and listings, so a model has a way to tell this Ordana apart from the others. The other was our own old listings, which are being rewritten to describe what Ordana does now. If you only take one thing from our check: before you write a new page, look at what the web already says about you. A stale listing can outrank you for your own name.

How do I increase my AI visibility (and get ChatGPT to recommend my SaaS)?

You influence the inputs, not the output. A search-grounded assistant builds its answer from the pages it finds for the question, so the job is to be on those pages, described the way you'd describe yourself. Your log's sources column tells you which pages those are. In rough order of where we'd start:

Your own page that answers the exact question

A page whose title is the buyer's question, with a two-or-three-sentence answer at the top that can be quoted on its own, then specifics: who it's for, how it's priced, what it doesn't do, how it compares. Say where a competitor is the better choice. Buyers trust it, and so does anything summarising you. Then make sure the page is actually indexed, in Google and in Bing: a site:yourdomain.com search in each takes a minute. ChatGPT's search has been widely reported to draw on Bing's index among others, so Bing is not optional.

The lists and directories that rank for the question

For "best X" questions, answers lean heavily on "best X" lists that already rank. In our own check, the results for the "best directories" questions were mostly lists like that. Your log shows which lists the assistants cite; ask each author to consider you, with an honest two-line pitch that says what category you're in. Directories whose pages rank count the same way, and the SaaS directories worth submitting to is our short list. If you're about to launch, the SaaS launch checklist puts the submissions in order.

One of those directories is ours. If one feature of your product works on its own, you can list it on Ordana as a widget for free: it gets its own public page, and other startups can add it to their product with no application or approval.

Review sites

For B2B category questions, review sites are often among the cited sources. A profile with no reviews does little, so set one up once customers will write one, and ask the happy ones.

Reddit and community threads

For the assistants whose search reads them, community threads are where opinions about your category form. Take part honestly: answer the question first, say the product is yours, link only when asked. How to promote your SaaS on Reddit without getting banned covers the method.

One name, one description, and sameAs

Use the same product name and the same one-line description everywhere: your site, every listing, every profile. If your name collides with another company's, pair it with your domain wherever you can. Add Organization structured data to your homepage with sameAs links to your own profiles, which ties them to your site. And go back to your old listings: they keep describing you long after you've changed.

llms.txt, as a minor helper

llms.txt is a proposed plain-text file at the root of your site that points AI tools at your most useful pages. It takes little time to write, and we have one. But as far as we know, none of the major assistant makers has said it decides what they recommend. Add it if you like; don't expect it to move your log.

Do I need a paid AI visibility tracker?

Paid AI-visibility trackers are their own product category now. They run large sets of prompts across several assistants on a schedule and report how often you're named against competitors, which sources are cited and how that changes over time. Free one-off "AI visibility checker" sites usually run a handful of their own prompts once, which is a quick snapshot.

The manual check is enough when:

  • You have one product, one market and one language.
  • A few dozen questions or fewer cover how your buyers ask.
  • Monthly is often enough, and you're the one who will act on the findings.

A tracker earns its fee when you need more prompts than a person can run, several markets or languages, weekly trend reports for a team, or a share-of-voice number against named competitors. It wins at volume and trends. The manual check wins at reading why: you see the whole answer, the wording and the sources, which is where the fixes come from.

How long does it take to show up in AI answers?

There's no fixed time, and anyone who quotes one is guessing. Search-grounded answers can change once the pages they read change and get re-crawled, which can be weeks or months. Answers from training data wait for the next model trained after your pages existed. That's the honest reason to log monthly: it's the only way to see which of your fixes landed.

Frequently asked questions

Is there a free AI visibility checker?

The most reliable free one is a spreadsheet and the four big assistants: put your buyers' real questions to ChatGPT, Claude, Gemini and Perplexity with web search on, more than once each, and log who gets named. Free one-off checker sites exist too, but they usually run their own prompts once, which is a snapshot, not a measurement.

How often should I check my AI visibility?

Once a month, with the same questions and the same settings, so the months compare. After a big change, such as a new comparison page, a fixed listing or a rename, recheck the questions it should affect.

Why does ChatGPT give a different answer every time?

Each answer is generated fresh. The exact wording, the search results at that moment and any personalisation all change it. That is why a single run tells you little: ask each question more than once and count how often you are named.

How do I increase my AI visibility?

Change what the answers are built from. Publish a page that answers each buyer question directly, get onto the lists, directories and review sites that rank for it, take part honestly in community threads, and use one consistent name and description everywhere, with Organization sameAs links to your own profiles.

Why does an AI assistant describe my product wrongly?

Usually because it is reading something old or someone else: an outdated listing, an old version of your site, or a different company with a similar name. Fix it at the source. Update the listings, keep one description everywhere and add sameAs links so your profiles are tied to your site.

Does llms.txt improve AI visibility?

Not much, as far as anyone can show. llms.txt is a proposed plain-text file that points AI tools at your most useful pages. It is cheap to add and does no harm, but as far as we know none of the major assistant makers has said it decides what they recommend.


Related reading: