How to measure your brand's AI visibility (without buying another tool)

October 2026 · 8 min read

For SaaS teams and agencies

Someone on your leadership team has asked whether ChatGPT recommends you. You probably tried it once, got an answer, and weren’t sure what it meant. Then you tried again and got a different one.

That’s normal. AI answers vary from run to run, by assistant, by location and by how the question is phrased. You can’t measure them the way you measure a Google ranking. But you can track them well enough to see direction, find the sources that shape the answers, and report something honest to the people who asked.

This is the method we’d use with a spreadsheet and two hours a month. It’s directional, not precise, and it’s worth saying so up front.

Step 1: Write 10–20 buying questions in your buyers’ words

The most common mistake is testing questions nobody asks. “What is the best AI-powered revenue intelligence platform?” is how a homepage talks. Buyers type something closer to “what do people use instead of Gong for a 10-person sales team”.

Good sources for real wording: sales call notes, the “how did you hear about us” field, support tickets from new customers, and the titles of Reddit threads in your category. Thread titles are useful because they show the exact phrasing people use when they ask peers.

Cover four types of question:

• Category: “best [category] for [situation]”
• Alternatives: “alternatives to [competitor]”, “[competitor] vs [competitor]”
• Problem: the pain in plain words, with no category named
• Fit: “is [your brand] good for [use case]”, to check how you’re described when you are named

Some examples of what that looks like in practice:

Fifteen questions is a good number. Fewer and one odd answer distorts the picture. More and the monthly check stops happening. Once chosen, freeze the list. Changing questions every month makes the trend meaningless. If you must add one, add it as a new row and keep the old ones.

Step 2: Pick the assistants and run them the same way every time

Check the assistants your buyers actually use. For most B2B teams that means ChatGPT, Perplexity, Google’s AI Overviews or AI Mode, and Gemini. Add Microsoft Copilot if you sell into enterprise IT. Three or four is plenty.

Consistency matters more than coverage. For every run:

• Use a clean session. Log out, use a private window, or use the assistant’s temporary or memory-off mode where it has one. Your own chat history and your company’s name in past conversations will bias the answer towards you.
• Fix the location. Answers change by country. If your buyers are in the US, run from the US or note that you didn’t. A UK team checking from London will see a different picture.
• Paste the question exactly. No follow-ups, no “what about [your brand]?”. You’re recording the first answer a buyer would get.
• Run each question twice. If the two answers disagree on whether you’re named, record both. That disagreement is information: it means you’re on the edge.
• Save the cited sources. Perplexity and Google’s AI features show their sources by default. ChatGPT and Gemini show them when they search the web. Copy the URLs, not just the domain names.

Same day each month, same person if possible. A screenshot of each answer takes seconds and settles arguments later.

Step 3: Score each answer on a simple sheet

Four fields are enough. Anything more and nobody fills it in.

• Mentioned: yes / no.
• Position: where you appear in the list of options (1st, 2nd, 5th), or blank if not named. First in a list of three is a different outcome from fifth in a list of eight.
• Sentiment: positive, neutral, negative or mixed. Judge the sentence about you, not the whole answer. “Cheaper, but harder to set up” is mixed.
• Cited sources: the URLs the assistant showed, especially any Reddit threads, review sites and comparison articles.

Here’s a template. The two filled rows are hypothetical examples to show the format, not real results.

At the end of each month, roll it up into three numbers per assistant: how many questions you were named in, your average position when named, and how many of those mentions were negative or mixed. Keep the raw rows. The summary tells you what changed, the rows tell you why.

Step 4: Read the sources, because that’s where the work is

The score tells you where you stand. The cited sources tell you what to do about it.

Put every cited URL from the month in one list and sort it by how often it appears. Patterns show up quickly:

• The same few Reddit threads appear again and again. Reddit was the most-cited domain across US AI search in Peec’s analysis of 30 million citations, so this is common. Open each thread. Are you mentioned? Is the top answer recommending a competitor? Is it two years old with nobody offering a more recent view?
• A comparison article you’re missing from. Note who wrote it and whether it’s been updated recently.
• Your own pages. If the assistant cites your site, check which page. If it’s a pricing or docs page rather than your homepage, that tells you which kind of page gets used.
• A source that’s wrong about you. Old pricing, a discontinued plan, a feature you’ve since shipped. Answers repeat what sources say.

Each thread or page you find becomes a line on a to-do list: a thread where an honest answer from someone who uses the product would help; an article whose author might take an update; a page on your own site that needs to answer one question directly. That list is the real output of the exercise.

If a negative thread keeps turning up in the citations, we’ve written separately about how to handle a negative Reddit thread that ranks for your brand. And if the threads are full of suspiciously identical praise for a competitor, see what to do about competitor astroturfing.

Step 5: Know what’s noise and what’s signal

Most month-to-month changes are noise. Treat these as noise:

• Appearing in one run and not the next for the same question.
• Moving one place up or down in a list.
• A single assistant changing while the others stay put, especially just after a model update.
• One new cited source that doesn’t show up again.

Treat these as signal:

• A change that holds for two or three months in a row.
• A change that shows up across several assistants at once.
• A change in how you’re described: the category you’re filed under, the use case you’re recommended for, or a repeated criticism.
• A new source that appears across several questions and assistants. Something on the web has started to matter.

The question to ask of any change is: can I point to a source that explains it? If the answers started naming you after a well-answered thread appeared in the citations, that’s a real result. If nothing in the sources changed, wait another month before telling anyone.

Step 6: Report it to leadership honestly

Leadership wants to know three things: are we showing up, is it getting better, and what are we doing about it. One slide covers it.

• Headline: one line with your real counts and the denominators visible. A hypothetical example: “Named in 6 of 15 buying questions on Perplexity, 4 of 15 on ChatGPT. Up from 4 and 3 last month.”
• What drives it: the three to five sources that appear most in the answers, and whether you’re present in each.
• What changed and why: only changes you can tie to a source. Say plainly when movement is probably noise.
• What’s next: the specific threads, articles and pages on the to-do list.

Add a line saying the method is directional: answers vary run to run, so trends over a quarter matter more than any single month. That sentence protects you when a number drops for no reason, and it will.

Resist turning this into a single “AI visibility score”. A blended percentage hides which assistant and which question moved, and that’s exactly what the team needs to see to act.

When a paid tool makes sense

The manual method works up to roughly 20 questions and four assistants. Beyond that, or if you need daily data, many prompt variations or several markets, a dedicated tool saves time. Products in this space include Peec AI, Profound and Otterly.AI, among others. Their coverage and pricing change often, so compare them against your own question list. Start with the manual version anyway: it teaches you which questions and sources matter, and that makes any tool you buy later more useful.

For the background on why assistants name some brands and not others, see why ChatGPT recommends your competitor and not you.

Sources: Peec, Top domains cited by AI search (March 2026) · Peec AI · Profound · Otterly.AI