The most reliable way to track whether ChatGPT recommends your brand is to run a fixed set of buyer questions on a schedule, across several AI engines, and log mentions, position and cited sources each time. A single manual check proves little, because AI answers vary from run to run. This guide gives you a repeatable weekly method, the metrics worth logging, and an honest view of when a paid platform is worth it.
Why is one ChatGPT check not enough?
Search-result summaries of several 2026 AI visibility guides make the same point: AI answers differ between runs, so one-off checks are weak and monitoring should involve repeated runs over time. A brand that appears once in a screenshot may be absent the next hour. What you want is a rate (how often you appear across many runs), not an anecdote.
The same guides note that classic SEO tools such as Ahrefs, Semrush and Moz track Google rankings, backlinks and keyword positions, not what ChatGPT says about your brand. So rank trackers you already pay for will not answer this question.
Sourcing note: we read these guides through search summaries, not the live pages, and most are written by vendors who sell AI visibility tools. Treat their rankings and statistics as marketing claims and verify any figure you plan to rely on.
Which metrics should you log?
The guides we reviewed list five metrics again and again. They are enough to start with:
| Metric | What it tells you | How to log it |
|---|---|---|
| Mention rate | In what share of answers your brand appears | Mentions divided by runs, per question and per engine |
| Position | Where you sit when the answer lists several brands | 1st, 2nd, 3rd mention |
| Share of voice | Your mentions compared with named rivals | Your mentions divided by all brand mentions |
| Sentiment | Whether the wording is positive, neutral or negative | A simple three-way label |
| Cited sources | Which pages the engine leaned on | The URLs listed with the answer |
Cited sources are the most actionable column. If the same review site or comparison page keeps being cited when a rival is recommended, that page is where you need to appear.
How do you build the prompt set?
Write the questions a buyer would type, not questions about your own name. A useful set has 15 to 30 prompts in three groups:
- Category questions: "best CRM for small agencies", "which scraper handles Cloudflare-protected sites".
- Comparison questions: "X vs Y for lead generation".
- Problem questions: "how do I monitor competitor prices every day".
Keep the wording fixed between weeks. If you change a prompt, you have changed the experiment and cannot compare the trend. Add your brand and three to five named competitors to the input so share of voice has something to divide by.
What is the weekly routine?
- Freeze the prompt set and the competitor list in a spreadsheet.
- Run every prompt on every engine you care about. The guides we read treat ChatGPT, Perplexity, Google AI Overviews, Gemini and Claude as the usual minimum.
- Repeat each prompt several times so you get a rate, not a single draw.
- Log the five metrics per prompt and engine.
- Compare with last week and read the cited sources for any prompt where a rival replaced you.
- Act on one thing: publish or update the page that the winning sources point to, or fix a crawl block. Our post on Cloudflare blocking AI crawlers covers the access side.
Doing this by hand across 20 prompts, 5 engines and several repeats means hundreds of answers a week. That is the point where automation pays for itself.
How can you automate it without a dedicated platform?
The AI Brand Visibility Tracker actor does the logging step. Per its own listing, it sends your questions to ChatGPT, Perplexity, Gemini, Claude and Grok and records, per answer, whether your brand is mentioned, its ranking position, sentiment, share of voice, which competitors appear and which sources are cited. It needs no OpenAI or Anthropic API key, and you pay per result on Apify with $5 of free credit for new accounts. Each row is one question on one engine, so you can filter and export to CSV, JSON or Excel.
Schedule it weekly on Apify and compare runs, or pass the dataset to n8n, Make or Zapier to get a message when your mention rate or position drops. If you want the raw model answers for your own analysis, the Bulk LLM Runner runs a spreadsheet of prompts across many models in parallel, and its price is per result.
For the wider toolkit, see the SEO and AI visibility hub.
When is a dedicated AI visibility platform worth paying for?
The guides list platforms such as Semrush and Profound among others. We have not tested them and do not quote their prices here, since the pricing pages were not readable in this research. A reasonable rule of thumb based on what the guides describe:
- Stay with a scheduled actor and a spreadsheet if you track one brand or a few clients, and you are happy reading a dataset.
- Consider a platform if you need dashboards for non-technical stakeholders, many brands, or features beyond mention tracking.
Either way, the method above does not change: fixed prompts, repeated runs, same metrics, weekly comparison.
What can you do once you see the data?
Tracking only matters if it changes what you publish. Three common moves follow from the cited-sources column:
- Get listed or reviewed on the sources engines cite for your category.
- Rewrite key pages answer-first, with a direct two-sentence answer under each question heading.
- Check crawler access and consider an llms.txt file, but measure before assuming it helps. We looked at that in Does llms.txt work?. For a first walkthrough, read Is ChatGPT recommending your competitors?
What mistakes make AI visibility data useless?
Most failed tracking projects share the same few errors. None of them need a tool to fix.
- Changing prompts between weeks. A reworded question is a new experiment. If you need new prompts, add them as a second set and keep the first one running, so the trend line stays intact.
- Only asking branded questions. "What is Acme?" tells you what the engine knows about you. It says nothing about whether you are recommended to someone who has never heard of you. Category and problem questions are where buyers discover brands.
- Drawing conclusions from one run. With answers that vary between runs, a drop from first to third position in a single sample may be noise. Wait for a pattern across several runs and two or three weeks before you react.
- Ignoring the engine mix. The guides we read note that coverage differs by engine, and that a tool focused on Google can have blind spots for ChatGPT and Claude. Log each engine separately instead of averaging them, so a gap on one is not hidden by a strong result on another.
- Tracking without a decision attached. Before the first run, write down what you will do if mention rate falls below a level you choose, for example refresh the page that the winning sources cite.
How do you read share of voice and sentiment?
Share of voice is simply your mentions divided by all brand mentions across the same prompts. If five brands are named in a set of answers and yours is named in a fifth of the mentions, that is your share. It is useful mainly as a comparison against the same rivals over time, not as an absolute number to quote publicly.
Sentiment is the rougher metric. A three-way positive, neutral or negative label is enough. What matters more is the wording: an answer that says your product is "cheaper but limited" is a mention, but it also hands you a specific message to address on your own pages. Read a sample of raw answers every month instead of trusting the labels alone.
Finally, keep a short log of what you changed and when: a new comparison page, a review site listing, a fixed crawler block. When the numbers move, that log is the only way to guess which change caused it, and you should still treat the link as a hypothesis, not proof.
Frequently asked questions
Weekly is a practical cadence for most brands. Run the same prompts each time so changes reflect the engine or your content, not your wording.
The guides we reviewed state that AI answers vary from run to run, which is why they recommend repeated runs over time instead of one-off checks.
According to the guides we read, traditional SEO tools track Google rankings, backlinks and keywords, not AI answers. You need a tool or process built for AI engines.
Not with the AI Brand Visibility Tracker actor. It runs without an AI API key and bills per result on Apify.
At least ChatGPT and Perplexity, plus Gemini and Claude if your buyers use them. The actor covers ChatGPT, Perplexity, Gemini, Claude and Grok.
Try it on your own brand
Run your first prompt set and see your mention rate today: start with the Apify actors or open the AI Brand Visibility Tracker directly.
Last updated: November 2026
β‘ Run this without building it
These actors already do what this guide describes. Free $5 credit on a new Apify account.
π Keep reading
Get the Free Web Scraping Toolkit
Join the newsletter and get my curated list of scraping tools, proxy comparison cheatsheet, and Python automation templates.