How to Measure Your Brand in AI Search for Free (Step-by-Step Method and Template)
This guide builds on our introduction to GEO. Research findings cited below come from studies of generative search published in 2025 and 2026, listed in that article's sources.
You cannot improve what you do not measure, and AI search is hard to measure. There is no Search Console for ChatGPT. Answers change from one run to the next. And most tracking tools that promise an "AI visibility score" charge a monthly fee to do something you can start doing yourself in an afternoon.
This guide gives you a free, repeatable method: a set of questions, a simple recording sheet, four metrics, and a monthly routine. It will not be as fast as a paid tool, but it will be honest, and it will tell you whether your GEO work is changing anything.
Why measuring AI visibility is tricky
Three findings from recent research shape the method:
- Answers vary between runs. Studies report substantial run-to-run variability: the same question can return different answers and different sources. One test proves nothing.
- Engines differ from each other. A 2026 study found very little overlap between the sources used by Google's classic results, AI Overviews and Gemini. Being visible in one engine says little about another.
- How you are described changes more than whether you are mentioned. A 2026 brand-tracking study found that whether a brand was framed positively or negatively flipped about 6.7 times more often than whether it was mentioned at all.
So the method below uses repeated runs, several engines and separate measures for mentions, citations and accuracy.
Set realistic expectations. The same brand-tracking study found small and niche brands appearing in about 11% of relevant AI answers on a first run, against 73% for household names. If you are a small business and start near zero, that is normal, not a failure.
Step 1: Write your prompt set
Write 15 to 30 questions your customers genuinely ask, phrased the way they would type or say them. Cover five types:
| Prompt type | Example (for a bookkeeping service) | What it tests |
|---|---|---|
| Category | "What are the best bookkeeping services for small businesses?" | Do you appear when people browse? |
| Problem | "How do I catch up on six months of messy books?" | Are you suggested as a solution? |
| Comparison | "Bookkeeping service vs accounting software for a freelancer?" | Are you included in trade-offs? |
| Alternatives | "Alternatives to [big competitor] for small businesses" | Do you appear next to competitors? |
| Brand | "What is [your brand]? Is it any good?" | Is the description accurate? |
Tips for good prompts:
- Use your customers' words, not your marketing language. Search Console queries, support emails and sales calls are good sources.
- Include your location if you serve a local market.
- Freeze the list. Use the same prompts every month so the trend means something. Add new ones in a separate section.
Step 2: Choose your engines and settings
Pick the three or four AI tools your audience actually uses, for example ChatGPT, Perplexity, Gemini and Google (for AI Overviews).
Keep the conditions as consistent as you can:
- Start a new chat for every prompt, so earlier questions do not influence the answer.
- Turn off personalisation or memory where possible, or use a logged-out or private session.
- Note your location and the date, because both affect results.
- Ask exactly the same wording each time.
Step 3: Run each prompt more than once
Because answers vary, run each prompt three times in each engine. That sounds like a lot, so do the arithmetic before you start: 20 prompts × 3 runs × 3 engines = 180 answers. At about 30 seconds each to run and record, that is roughly 90 minutes a month. If that is too much, cut to 15 prompts or two engines, but keep the repeats.
Step 4: Record six things for every answer
Use a spreadsheet with one row per answer and these columns:
| Column | What to record |
|---|---|
| Date, engine, prompt ID, run number | So you can compare like with like |
| Mentioned? | Yes or no: is your brand named anywhere in the answer? |
| Position | If mentioned, first, early or late in the answer |
| Cited? | Yes or no: is there a link to your website? |
| Cited URL | Which page, if any |
| Accurate? | Accurate, partly wrong or wrong: pricing, location, what you do |
| Competitors named | List them, for share of voice |
| Sources cited | The other sites the answer relied on |
The "sources cited" column is the most useful one for action: it shows you which third-party sites AI engines rely on in your space. Those are the places where being mentioned is most likely to help.
Step 5: Calculate four metrics
1. Mention rate = answers mentioning you ÷ total answers. 2. Citation rate = answers linking to your site ÷ total answers. 3. Share of voice = your mentions ÷ (your mentions + mentions of your main competitors). 4. Accuracy rate = accurate descriptions ÷ answers that mention you.
Illustrative example: you run 180 answers and are mentioned in 27. Your mention rate is 27 ÷ 180 = 15%. If 9 of those include a link, your citation rate is 9 ÷ 180 = 5%. If competitors are mentioned 63 times in the same answers, your share of voice is 27 ÷ (27 + 63) = 30%. If 24 of your 27 mentions describe you correctly, accuracy is 24 ÷ 27 ≈ 89%.
Calculate each metric per engine as well as overall, since engines behave differently.
Step 6: Add the traffic side
AI visibility only matters if it reaches people. Two free checks help:
- Analytics referrals. In your analytics tool, look at referral traffic from AI tools, for example from domains such as chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. Track sessions and conversions from them monthly.
- Search Console. Watch impressions and clicks for the questions in your prompt set. Search Console may not separate AI Overview traffic from regular results, so treat it as a broad signal.
Ask new customers "how did you hear about us?" and include "ChatGPT or another AI tool" as an option. It is the simplest attribution method there is, and often the most revealing.
Step 7: Turn the data into actions
Look at the sheet each month and ask four questions:
- Where am I never mentioned? Those prompts show content gaps. Do you have a page that clearly answers that question?
- Which sources keep getting cited instead of me? Those are the review sites, lists, communities and publications where you should earn a presence.
- What do engines get wrong about me? Fix the facts on your own site first, then on your profiles and listings. Inconsistent information is a common cause.
- Which pages get cited? Study them. They usually answer directly, include specific facts and are well structured. Do more of that.
Then make one or two changes per month and keep measuring. Because results vary, judge changes over two or three months, not one.
A monthly routine
| When | What | Time |
|---|---|---|
| First week of each month | Run the prompt set, record answers | 60–90 minutes |
| Same day | Update the four metrics per engine | 15 minutes |
| Same day | Check AI referral traffic and "how did you hear about us?" answers | 10 minutes |
| Following week | Choose and make one or two improvements | Varies |
| Every quarter | Review trends; retire or add prompts in a separate section | 30 minutes |
When a paid tool is worth it
Doing this by hand is fine for 15 to 30 prompts. Paid AI-visibility trackers start to make sense when you need hundreds of prompts, many locations or languages, daily tracking, or reports for clients. Before paying, check how a tool handles the variability problem: does it run each prompt several times, and does it separate mentions, citations and accuracy? A single "visibility score" can hide all three. Some SEO tools now include AI-visibility tracking; our Semrush alternatives guide and SEO lifetime deals roundup compare options, and our 30-minute evaluation method helps you test one before committing.
Common mistakes
- Testing once and drawing conclusions. Variability makes single runs meaningless.
- Changing the prompt list every month. Trends need a fixed baseline.
- Only tracking mentions. A mention with wrong facts can do more harm than no mention.
- Testing while logged in with history on. Your own past chats can shape answers.
- Chasing every engine. Focus on the ones your customers use.
💡 Keep learning
What is GEO? → · SEO lifetime deals → · Semrush alternatives →
Frequently asked questions
How do I check if ChatGPT mentions my brand? Write a fixed list of real customer questions, run each one several times in a new chat with memory off, and record whether your brand is mentioned, cited with a link, and described accurately.
Why do I get different answers each time? AI answers are generated fresh and can draw on different sources each time. Research shows substantial run-to-run variability, which is why you should run each prompt several times and look at trends.
What metrics should I track for AI search visibility? Mention rate, citation rate, share of voice against competitors, and accuracy rate, ideally per engine.
How often should I measure? Monthly is enough for most small businesses. Judge changes over two or three months, because individual results vary.
Can I see AI search traffic in Google Analytics? You can look at referral traffic from AI tools such as chatgpt.com and perplexity.ai. Asking new customers how they heard about you adds a simple, reliable signal.
Do I need a paid AI visibility tool? Not at first. A spreadsheet and 15 to 30 prompts are enough. Paid tools help when you need large prompt sets, many locations or client reporting.