AI Search and GEO · How to
Track ChatGPT mentions for free: the weekly 10 prompt log
Part of the guide Generative engine optimization (GEO): a practical guide for 2026
You can track ChatGPT mentions for free with a spreadsheet and about 20 minutes a week per engine. Write 10 prompts the way a buyer types them, run each one logged out three times, ignore the answer text, open the sources, and log five things: the date, the engine, whether you were named, which URLs were cited, and which competitors were named. Read the trend monthly.
Why a manual log beats guessing
A manual log shows you what a logged-out buyer actually sees, including the sources behind the answer. Most paid trackers query engines through their APIs, and API answers can cite different sources than the consumer app. A small, consistent manual sample is the cheapest way to see the real thing. The method below follows the weekly prompt log from the GEO guide by Nicholas Dulait (November 2026), which we also use for Postbing. It is one of the four measurement methods in the generative engine optimization guide.
How this fits with GA4, Search Console and paid tools is covered in the AI visibility tracking guide. This page is the hands-on part.
Step 1: Write your 10 prompts
Write prompts the way a buyer types them late at night, not the way a marketer names the category. Include a situation and a constraint. Compare:
| Marketer phrasing (weak) | Buyer phrasing (strong) |
|---|---|
| best CRM software | what's a simple CRM for a 2 person roofing business that works on my phone |
| project management tools | cheapest alternative to [competitor] for a small design studio |
| email marketing platform | is [your brand] or [competitor] better if I only send one newsletter a month |
A balanced set of 10: four category prompts with a constraint, two "alternative to [competitor]" prompts, two "[A] or [B]" prompts, one "is [your brand] any good", and one prompt about a problem your product solves. Freeze the list for a quarter so the trend means something.
Step 2: Set up the sheet
Create one tab with these columns. Copy the header row as is:
| Date | Engine | Prompt # | Run | Named? (Y/N) | Recommended? (Y/N) | Our URLs cited | Other URLs cited | Competitors named | Notes |
|---|---|---|---|---|---|---|---|---|---|
| 2026-11-16 | ChatGPT | 3 | 1 | Y | N | yoursite.com/vs/rival | reviewsite.example/best-crm | Rival A, Rival B | Named third of four |
| 2026-11-16 | ChatGPT | 3 | 2 | N | N | reviewsite.example/best-crm, forum.example/thread | Rival A | ||
| 2026-11-16 | Perplexity | 3 | 1 | Y | Y | yoursite.com/pricing | Rival B | Quoted our price |
The rows above show the format with placeholder sites, not real results. Put the prompt text on a second tab and refer to it by number, so the log stays readable.
Step 3: Run each prompt correctly
Consistency matters more than volume. For every prompt, in every engine you track:
- Log out, or use a private window. In ChatGPT a temporary chat is the closest option if you cannot log out. Logged-in answers are shaped by your history.
- Paste the prompt exactly as written. Do not tweak wording between weeks.
- Make sure the engine searched. You want an answer with sources. If none appear, note "no search" and move on.
- Open the sources. This is the step everyone skips, and it is the useful part. The sources tell you which pages and page types win the question.
- Repeat twice more in a fresh window. Three runs per prompt per engine, because the same prompt returns different sources.
- Log immediately, before you open the next run.
Ten prompts times three runs is 30 answers per engine. Start with the engine your buyers use most and add a second one when the routine is easy.
Step 4: Add the formulas
Add a summary tab. In Google Sheets, with the log on a tab named Log and columns laid out as above (Engine in B, Named in E, Recommended in F, Our URLs in G):
- Named rate for ChatGPT:
=COUNTIFS(Log!B:B,"ChatGPT",Log!E:E,"Y")/COUNTIF(Log!B:B,"ChatGPT") - Recommended rate: the same formula with column F.
- Cited rate:
=COUNTIFS(Log!B:B,"ChatGPT",Log!G:G,"<>")/COUNTIF(Log!B:B,"ChatGPT") - Competitor share:
=COUNTIFS(Log!B:B,"ChatGPT",Log!I:I,"*Rival A*")/COUNTIF(Log!B:B,"ChatGPT")
Add a date filter (a COUNTIFS condition on column A) to get weekly and monthly numbers. Chart the monthly named rate and cited rate per engine. That is your whole dashboard.
Step 5: Read the results
Look at three things once a month, not every week.
Which URLs keep showing up. List every third-party URL cited in at least three runs. These pages are where the answers come from. If a competitor appears on three of them and you appear on none, that is your main problem, and it is not on your own site.
Which of your pages get cited. If only your home page or pricing page appears, you are missing pages that match the question's shape, such as comparisons or alternatives. How to get cited by ChatGPT covers the fixes.
Whether named and recommended move together. Being named but never recommended usually means the pages describing you do not tie you to a clear use case. How to rank in ChatGPT search covers positioning.
Step 6: Turn the log into one action a week
End each session with a single action tied to a row: email the author of a list that cites three competitors and not you, fix the outdated price a page was quoted for, or write the comparison page that keeps being cited from a competitor's site. One action a week compounds; a long list does not get done.
Re-run the same prompts next week. Judge changes over at least two months, because sources rotate on their own and a single good week is often noise.
What this method cannot do
It is a sample of 30 answers per engine, so it will miss rare prompts and cannot estimate total traffic. It does not see logged-in, personalised answers. It does not show Google AI Overview impressions at scale; use Search Console for that. Pair it with the GA4 AI channel to see which mentions turn into visits.
If you would rather automate the weekly runs, Postbing includes an API-based AI visibility check in every plan, with 5 to 40 questions per site per week depending on plan, as of November 2026. Because it uses APIs, keep doing a manual logged-out check at least once a month. See Postbing pricing.
FAQ
How long does the weekly log take?
About 20 minutes per engine for 10 prompts and three runs each, once you have the sheet set up. The first week takes longer because you are writing prompts.
Why run each prompt three times?
Because AI engines rotate their sources, and the same prompt can name you in one run and not the next. Three runs per prompt gives a rate instead of a yes or no.
Should I use my own ChatGPT account?
No. Log out, use a private window, or use a temporary chat. Logged-in answers are shaped by your history and will flatter you.
Can I track Google AI Overviews the same way?
Yes, by searching your prompts in an incognito window and logging whether an AI Overview appears and who it cites. For volume, use Search Console, which includes AI feature traffic in its Performance report.
Sources
- Nicholas Dulait, "The KING of GEO" post on X (13 November 2026): weekly 10 prompt method, logged-out runs, three runs per engine, opening sources, API vs app caveat, accessed 16 November 2026
- Google: AI features and your website: AI feature traffic in the Search Console Performance report, accessed 16 November 2026
Related reading
Last updated: November 16, 2026. Spotted an error? Email support@postbing.com and we will fix it. Editorial policy.
Try Postbing free
Postbing reads your site, plans a month of topics from what your customers search for, writes in your voice and publishes to WordPress, Shopify, Webflow or Ghost. Your first article is free.
Try Postbing freeNo card needed. No link exchanges, ever.