By Raluca Baciu, Vias Digital. Published 8 August 2026.
To monitor brand mentions in ChatGPT, run a fixed battery of 10 to 15 buyer questions through it once a month, screenshot and log every answer, benchmark three competitors with the same questions, and fix the sources the wrong answers come from. It costs nothing but an hour a month, and it produces the one dataset most leadership teams have never seen: what the AI your buyers consult actually says about you. Here is the exact five-step method, the same one we run for clients at Vias Digital.
Because it is where your buyers start. ChatGPT is the default research assistant for most B2B buying committees, and its answers increasingly shape shortlists before any website visit happens. Gemini, Perplexity, Claude and Copilot matter too, and our guide to tracking brand mentions across AI search covers all five. But if you only monitor one system this quarter, make it this one.
Step 1: Build your question battery. Write down the 10 to 15 questions your buyers ask before they shortlist a vendor, phrased the way they would actually type them. Cover five categories: category recommendation (who should I choose for X), direct comparison (you vs named competitors), credibility (is company X legitimate, who is behind it), pricing and process (what does X cost, how does working with X look), and hiring (is X a good place to work). Keep the wording fixed forever: the battery is your measuring instrument, and changing the questions resets the trend.
Step 2: Run it on a fixed schedule. Once a month, same conditions every time: a fresh chat, browsing on if your buyers use it, the exact same question wording, one question per conversation so answers do not bleed into each other. Note the date and the model version. Consistency is what turns anecdotes into data.
Step 3: Record the answers as evidence. Screenshot every answer, then log it in a spreadsheet with six columns: the question, the date, were you mentioned (yes or no), how you were framed (one line), which competitors were named, and which sources ChatGPT cited. After three months you have a trend line most companies have never seen.
Step 4: Benchmark your competitors with the same battery. Run the identical questions for the three competitors you actually lose deals to. The point is not their score, it is the delta: the questions where they appear and you do not are your highest-priority gaps, and the sources cited for them are your fix list.
Step 5: Fix the sources ChatGPT reads. ChatGPT forms its picture of you from what it can read: your website, your LinkedIn company page and executive profiles, directories, review platforms, press coverage. Correct factual errors at the source, publish answer-first content that states plainly who you are and what you do, and close the source gaps where competitors are cited and you are absent. Then verify the fix in next month's run.
Manual monitoring is honest work, and it is enough for one market and one language. It stops being enough when you need five AI systems instead of one, fifty questions instead of fifteen, several markets and languages, alerting when an answer changes, or a report your board will actually read. At that point the choice is between a monitoring tool, which automates the collection and leaves the interpretation to you, and a managed read like the AI Narrative Monitor, which delivers the interpretation itself: quarterly, sourced, with the three fixes that matter most, ranked.
How to Track Brand Mentions in AI Search, the full picture across ChatGPT, Gemini, Perplexity, Claude and Copilot, with the leading tools compared.
Brand Visibility in AI Search, what the metric means and a scorecard for measuring yours.
Monthly is the right rhythm for ChatGPT. Answers shift with model updates and browsing behaviour, and a month is short enough to catch a narrative problem before a buyer acts on it. Quarterly is the minimum viable cadence.
No. Answers vary with phrasing, conversation history, location and whether browsing is on. That is why the playbook uses a fixed question battery run under the same conditions each time: you are tracking the trend, not any single answer.
No. There is no paid placement inside ChatGPT answers. What you can influence are the sources ChatGPT reads and trusts: your own site, your LinkedIn presence, directories, reviews and earned media. That is the entire game.
Trace the claim to its source. Wrong answers almost always come from a specific page, directory listing or outdated article the model trusts. Correct the source, publish the accurate version in answer-first formats on your own properties, and verify the fix in your next monthly run.
Manual is enough to start and to run for a single market. You outgrow it when you need several AI systems, larger prompt sets, alerting, or board-ready reporting. At that point, compare the monitoring tools or use a managed read like the AI Narrative Monitor.