How to Track Brand Mentions Across AI Platforms Using a Repeatable Test Set
To track brand mentions across AI platforms reliably, you need the same discipline you would use for any experiment: fixed inputs, consistent conditions and a clear way to record results. Typing a few questions into ChatGPT whenever someone is curious produces anecdotes, not data. A repeatable test set turns those anecdotes into a trend you can act on. This guide walks through building the prompt set, running it, scoring the answers and comparing results over time.
Why Ad Hoc Checks Fail
AI answers change with wording, location, account history and model updates. If one person asks “best SEO agency in Delhi” and another asks “top SEO company Delhi”, they may get different brands. Neither result tells you whether your visibility improved. Only the same questions, asked the same way, on a regular schedule, can show genuine movement. This is also why so many teams wrongly conclude that their brand is not showing in ChatGPT after a single test.
Step 1: Build the Prompt Test Set
Start From Real Customer Questions
Collect questions from sales calls, support tickets, site search, Search Console queries and People Also Ask boxes. Rewrite them the way people talk to an assistant, which tends to be longer and more specific than a search query. Our work on conversational SEO uses the same source material.
Cover Every Stage of Intent
Balance the set across these prompt types:
| Prompt type | Example |
|---|---|
| Category discovery | What are the best AI SEO agencies in India? |
| Problem-led | Why is my website not appearing in ChatGPT answers? |
| Comparison | SEO Plus GEO vs a traditional SEO agency |
| Brand direct | What does SEO Plus GEO offer? |
| Local | Who offers generative engine optimization in Noida? |
Branded prompts tell you whether AI describes you accurately. Unbranded prompts tell you whether AI recommends you at all. You need both.
Keep the Set Stable
Assign each prompt an ID, a topic and an intent label. Freeze the wording. If you must change a prompt, retire the old ID and create a new one so history stays clean.
Step 2: Control the Test Conditions
Record these settings and keep them constant:
- Platform and model version where visible
- Logged-out or a dedicated clean account with memory turned off
- Country and language, using the same location every run
- Search or browsing mode switched on or off consistently
- Date and time of the run
Because answers vary, run each prompt three to five times per cycle. Treat a brand as “mentioned” if it appears in the majority of runs, and keep the raw counts for anyone who wants detail.
Step 3: Score Every Answer
Use a simple, consistent scoring sheet. For each run, capture:
- Mentioned: yes or no
- Position: first, within the top three, or later
- Cited: whether your domain appears as a linked source
- Cited URL: which page was linked
- Sentiment and accuracy: positive, neutral, negative, and whether facts are correct
- Competitors named: every other brand in the answer
Accuracy matters as much as presence. If an assistant mentions you with outdated pricing or wrong locations, that is a problem to fix on your own website first, where most assistants get their facts.
Save the Evidence
Store the full answer text and the source list for every run. When someone asks why the mention rate fell, you can show the answers rather than relying on memory.
Step 4: Track Brand Mentions Across AI Platforms Over Time
Run the full set on a fixed cadence, typically every two weeks or monthly. Then calculate:
- Mention rate per platform and overall
- Citation rate per platform
- Share of voice against named competitors
- Accuracy rate for branded prompts
Plot each metric over time. Add notes when you publish major content, earn press coverage or notice a model update, so the team can connect changes to likely causes. Our AI brand mention case study shows how this annotated timeline made results easy to explain.
Turning Results Into Actions
The test set is valuable only if it changes what you do. Typical patterns and responses:
- Mentioned but rarely cited: create a definitive page on the topic that is easy to quote, and strengthen entity SEO optimization so platforms connect the page to your brand.
- Cited but described inaccurately: fix inconsistent facts across your site and profiles.
- Absent on one platform only: check platform-specific sources. Gemini leans on Google’s index, while Perplexity performs live retrieval, so the fixes differ. See our guides on Perplexity visibility ranking and Gemini visibility ranking.
- Competitors dominate comparisons: publish fair comparison content and earn mentions on third-party sites that assistants already trust.
Tools Versus Manual Testing
Monitoring tools can run prompts at scale and save time, but they make different methodological choices, which is why their numbers rarely match. A manual or semi-automated test set gives you full control over wording and conditions. Many teams combine both: a tool for breadth and a smaller manual set for accuracy checks.
Whichever route you choose, store results in one sheet with the same columns, so tool data and manual data can be compared side by side. Agree on a single definition of “mentioned” before the first run. For example, decide whether a brand named only in a source list counts, or only a brand named in the answer text. These small definitions change totals more than most people expect. If you want the monitoring connected to a wider plan, our AI search ranking work and approach to how AI chooses sources build directly on this data.
Frequently Asked Questions
How Many Platforms Should I Test?
Start with ChatGPT, Google AI Mode or AI Overviews, Gemini and Perplexity. Add Copilot or Claude if your audience uses them.
Should I Test Logged In or Logged Out?
Use a clean account with memory and personalization disabled, or logged-out sessions where possible, and never mix the two within one cycle.
How Long Before Changes Show Up?
Live retrieval platforms can reflect new content within days. Changes that depend on model training can take months, so judge progress over several cycles.
A repeatable test set is the most dependable foundation for AI visibility work. To build one for your brand, explore our ChatGPT visibility ranking service, learn how to get mentioned in ChatGPT answers, or review our LLM optimization approach.