How to Check What AI Says About Your Brand: A Free 30-Minute Audit

What's the fastest way to check what AI says about your brand?

Open a fresh, signed-out chat on two or more AI engines. Ask 5-8 buyer-style questions in your category, without naming your brand, three times each, in separate chats. For every answer, log whether your brand was named, described correctly, recommended, and cited — plus whatever source got cited instead, if it wasn't you. That's the whole manual routine: roughly 30 minutes once you've written the prompts, and it costs nothing but time.

The rest of this post breaks that routine into the five decisions that actually determine whether it's worth anything: which prompts, how many runs, how to log and score, which free tools can add real traffic data behind the chat answers, and where a manual pass runs out of road.

Which prompts should you start with?

Write prompts the way a prospect would type them, not the way your marketing page talks about you:

• 3-4 need-based prompts with no brand name: "What's a good tool for [the job your product does]?" or "How do I [solve the problem your product solves]?"
• 2 comparison prompts: "What are the best options for [category] in [year/market]?" or "[Competitor] vs. [another competitor] — which is better for [use case]?"
• 1-2 fact-check prompts that do name you: "What does [your brand] do?" or "Is [your brand] good for [use case]?" These check accuracy, not just visibility.

Skip prompts that only a current customer would ask ("How do I reset my [product] password?"). Those test support content, not whether AI recommends you to someone who hasn't heard of you yet.

How many runs per prompt — and why one isn't enough

Run each prompt at least three times per engine, in separate fresh chats, not as follow-ups in the same thread. A single run tells you what happened once; AI answers vary between runs on the same prompt, sometimes by a lot, so a brand that's missing on run one can appear on run two. Three runs is a reasonable floor for a manual check — enough to notice if an engine is inconsistent, not enough to call it a statistically reliable rate. For what repeated, scheduled runs across more prompts look like at a larger scale, see how to measure AI visibility.

How to log and score the answers

Use one row per prompt, per engine, per run. Score four columns against your own facts — price, features, who you serve, where you operate — not against how the engine phrases things:

Fill in one row per run. A prompt run three times on two engines is six rows. After a dozen or so prompts, patterns show up fast: which claims are wrong, which competitor's content keeps getting cited instead of yours, and whether you're named but never recommended — three different problems with three different fixes.

Which free tools add real data behind the chat answers?

Manually asking chats only shows you a sample. Three free, first-party reporting surfaces show actual traffic and citation data, not a simulated question:

• Bing Webmaster Tools' AI Performance report (public preview, announced by Microsoft in February 2026) shows how your site's pages are cited as sources in Copilot and Bing's AI-generated answers — total citations, average cited pages, grounding queries, and which specific URLs got referenced. It requires a verified site in Bing Webmaster Tools.
• Google Search Console's generative AI performance report shows impressions for your pages inside Google's AI Overviews and AI Mode, broken down by page, country, device, and date. As of August 31, 2026, Google says this report has rolled out to all sites worldwide, though it only populates once a site has enough AI-feature impressions to report.
• GA4's "AI Assistant" channel, inside the Default Channel Group in your acquisition reports, automatically buckets sessions with the medium ai-assistant when a visitor arrives via a recognized AI referrer. Google's own Default Channel Group documentation lists ChatGPT, Gemini, DeepSeek, Copilot, and Grok as recognized sources — notably not Claude or Perplexity, so traffic from those still lands in Direct or Referral. [VERIFY] before you rely on it: check your own GA4 property's channel list, since Google has updated the recognized-referrer list before.

None of these three tools run a prompt for you. They tell you what already happened — who clicked through from an AI answer, and which of your pages got cited — which is the half of the picture a manual chat session can't see.

ContextOwl's free tools for the parts this routine doesn't cover

Six free tools handle pieces of this check without a login: the AI Citation Snapshot runs one prompt on one engine and shows whether you're named and cited; Competitor AI Comparison runs the same prompt against a named competitor; the AI Bot Access Checker confirms AI crawlers can actually reach your site; the llms.txt Generator builds the file some engines use to understand a site's structure; the Direct-Answer Snippet Checker flags whether a page answers a question in a citable, extractable format; and the FAQ Schema (JSON-LD) Generator builds structured FAQ markup. None of these replace the manual logging sheet above — they each answer one narrower question faster.

When does a manual check stop being enough?

A 30-minute pass is enough to catch an obviously wrong fact or spot that a competitor's page gets cited and yours doesn't. It stops being enough once you need any of the following, which a spreadsheet and a free hour can't deliver: the same prompt set checked on a recurring schedule, enough prompts and engines to call a result a rate rather than a sample, or a record you can hand to a stakeholder as evidence rather than a screenshot. At that point, a structured audit or a one-time report does the repeating and scoring for you — see the AI search audit checklist for the fuller version of this process, or how to choose an AI visibility tool if you're evaluating paid options.

FAQ

Do I need a paid tool to check what ChatGPT says about my brand? No. The manual routine in this post — prompts, three runs per engine, a logging sheet — is free and takes about 30 minutes once the prompts are written.

Should I include my brand name in every prompt? No. Most prompts should describe the need or category without naming your brand, since that's how a real prospect asks. Reserve a couple of prompts that do name your brand, to check whether the facts AI states about you are accurate.

What counts as a "wrong" answer versus a "missing" one? Missing means the engine didn't name you at all for a prompt where a competitor appeared. Wrong means it named you but got a fact wrong — price, features, location, or who you serve. Each needs a different fix: missing means finding what gets cited instead, wrong means fixing the claim at its source.

Why run each prompt three times instead of once? Because AI answers to the same prompt can change between runs. One run only shows what happened that one time; three gives you a rough sense of whether an answer is consistent or a coin flip.

Does GA4's AI Assistant channel catch all AI-referred traffic? Not all of it. It's built around a specific list of recognized referrers, and as of this writing that list doesn't include Claude or Perplexity on Google's own Default Channel Group help page — so some AI-referred sessions still show up as Direct or Referral. [VERIFY] your own property's current list before reporting on it.

What to do next

1. Write 5-8 prompts (mostly unbranded, a couple branded) and set up the logging sheet above.
2. Run each prompt three times per engine, in fresh chats, and score named, accurate, recommended, and cited for every row.
3. Check your Bing Webmaster Tools AI Performance report and Search Console's generative AI report for citation and impression data that backs up what you logged.

This manual routine is a starting point, not a substitute for a structured audit; if you want a faster first look at one prompt on one engine, try the AI Citation Snapshot.