How to measure AI visibility without paid tools

AI visibility tools cost anywhere from a modest monthly fee to enterprise contracts. Before you pay for one, it helps to know what you're measuring. The good news: you can do a solid version by hand, for free, in about an hour a fortnight.

This is the method we'd use on a new product. It combines a fixed set of buyer prompts, a simple scoring sheet, and the free official data that Google, Microsoft and OpenAI now provide.

Short answer: Write 20–30 prompts that real buyers ask. Every two weeks, run each one in ChatGPT, Perplexity and Google in a clean session, at least twice. Record whether you're mentioned, whether you're linked, and who else appears. Track mention rate and citation rate over time, and cross-check with Search Console's Generative AI report, Bing's AI Performance report, and ChatGPT referrals in analytics.

What "AI visibility" actually means

It's worth separating three things, because they move differently:

1. Mention: the answer names your product.
2. Citation: the answer links to one of your pages as a source.
3. Recommendation: the answer suggests your product for the user's need, ideally with accurate facts.

You can be mentioned without being cited, for example when a model knows your name from training data. You can be cited without being recommended, for example when your blog post is used as a source for a general answer. Track all three.

Our glossary has short definitions of LLM citation and AI share of voice if you're sharing this with a team.

Step 1: write your prompt set

Aim for 20–30 prompts, in five groups. Write them the way a buyer would type them into a chat, not the way you'd type them into Google.

Category (6–8 prompts). "What's the best tool for [job] for [audience]?" "What software do [audience] use for [job]?"

Comparison (4–6). "[Competitor] vs [you]: which is better for [use case]?" "Compare [competitor A] and [competitor B]."

Alternatives (4–6). "What are good alternatives to [competitor]?" "Cheaper alternative to [competitor] for [audience]?"

Problem (4–6). "How do I [problem your product solves]?" These test whether your guides get cited even when the user isn't shopping.

Brand (2–4). "What is [your product]?" "How much does [your product] cost?" These test whether the facts about you are right.

Keep the prompt set fixed. If you change the wording every round, you can't compare rounds. Add new prompts in a separate section, and only compare like with like.

Step 2: control the conditions

AI answers vary. OpenAI's help article on ChatGPT search explains that ChatGPT rewrites your question into search queries, may run follow-up queries, and may use saved memories when rewriting. So a signed-in account that has talked about your product for months is the worst possible test environment.

To reduce noise:

• Use a clean session. A logged-out or fresh account, or a private window, with memory and personalization off where the product allows.
• Keep location constant. Answers can differ by country. Test from the market you sell in.
• Force a web search when you can. In ChatGPT, use search mode, so you're testing cited answers, not only training memory.
• Run each prompt at least twice per round. Record both. A mention in one run out of two is different from two out of two.
• Test on the same day each round. Every other Monday, for example.

For Google, run the query in normal Search and note whether an AI Overview appears and whether you're linked in it. Try AI Mode for the category and comparison prompts.

Step 3: score each answer

A spreadsheet with one row per prompt, per platform, per run. Columns:

The "Sources" column is the most useful one. It tells you which pages the AI trusts for your category: a directory, a comparison site, a forum thread, a competitor's blog. Those are the pages you want to be on, or to outdo.

Step 4: turn it into four numbers

Per platform, per round:

• Mention rate: runs where you're mentioned ÷ total runs.
• Citation rate: runs where one of your URLs is cited ÷ total runs.
• Share of voice: your mentions ÷ all product mentions in the answers, across you and your tracked competitors.
• Accuracy rate: runs with no factual errors about you ÷ runs where you're mentioned.

Plot them over rounds. With 25 prompts, 3 or 4 platforms and 2 runs each, you'll have 150 to 200 answers per round. That's enough for a trend, not for decimals. Treat a change of a few points as noise until it holds for two or three rounds.

Step 5: add the free official data

Manual checks tell you what answers look like. Official data tells you what happened at scale. Three sources are free.

Google Search Console. The Generative AI performance report, rolled out to all sites on August 31, 2026, shows impressions from AI Overviews and AI Mode by page, country, device and date. No clicks or queries yet, but it tells you which pages Google's AI features show. We cover it in AI Overviews and AI Mode for founders.

Bing Webmaster Tools. The AI Performance report shows total citations in Microsoft Copilot and Bing AI summaries, which of your pages are cited, and the grounding queries the AI used. Since June 2026 it also shows citation share: "the percentage of citations attributed to your site out of all citations shown across all sites for that same grounding query." That's share of voice, measured by the vendor.

Your analytics. OpenAI's publisher FAQ says ChatGPT adds utm_source=chatgpt.com to referral links. Filter by it. Look for referrals from perplexity.ai too. Those are the visits AI answers actually sent you.

Step 6: act on what you find

Measurement is only useful if it changes what you do. Common findings, and what they point to:

Not mentioned anywhere. Check the basics first: are AI search crawlers allowed? Run the AI crawler checker. Then look at your "Sources" column. If AI answers cite directories and comparison pages you're not on, get on them. Our directory list and a free launch are starting points.

Mentioned, never cited. The model knows you, but your pages aren't what it retrieves. Write pages that answer the search-style version of the prompt directly, near the top. Our guide to getting cited by ChatGPT and Perplexity goes through it.

Cited, but facts are wrong. Usually old pricing or an outdated description somewhere. Fix your own pages first, then update the listings you control. Consistent facts across the web are the fix.

Strong on one platform, absent on another. Different products search differently. If you show up in Copilot but not ChatGPT, check your Bing and OAI-SearchBot status. If you're in Perplexity but not Google's AI features, look at your Google indexing and titles.

When the manual method stops being enough

The spreadsheet works well up to about 30 prompts and one product. It gets painful when you want weekly rounds, more prompts, several markets or several products. That's the point to consider software.

We compare options on best AI visibility trackers. Disclosure: our own Autopilot, in early access at $49/mo, includes AI-visibility tracking for 25 prompts across ChatGPT, Perplexity, Gemini and Claude, tied to the content and Search Console work that's meant to move those numbers.

Whichever you pick, keep the method: a fixed prompt set, clean conditions, repeated runs, and trends over single answers. A tool that doesn't tell you how it controls for those is giving you screenshots, not measurement.