Free template
You know your rankings. You have no idea what AI says about you.
Ask ChatGPT who the best plumber in Scarborough is and it names three businesses. Nothing in your analytics tells you whether you were one of them. This is the tracker we use with clients to find out.
No email. No signup. Rebrand it and use it with your own clients if you like.
Prompt
Logged as
Same business, same question, four engines, four different answers. The redacted names are competitors. This is the surface your rank tracker cannot see.
The problem
Rank tracking does not cover this.
Rankings and AI answers move independently. We have clients at position three for a term who never get named by ChatGPT for the question behind it, and clients recommended by Perplexity off a single Reddit thread while ranking nowhere.
They answer questions, not keywords
Nobody types “SEO agency Toronto” into ChatGPT. They type “I run a dental clinic in Toronto and I need help with marketing, who should I talk to?” Different question, different answer set.
They cite different sources
Perplexity leans on the open web and shows its work. AI Overviews leans on Google’s index. ChatGPT leans on its training data plus Bing. The same business can be invisible in one and dominant in another.
The answer is a shortlist
Ten blue links gave you room to be sixth. An assistant names three. Sixth does not exist.
There is no proxy metric that gets you there. You have to measure it directly, which is what the spreadsheet is for.
The file
Four tabs. One of them you type in.
Visibility Log
One row per prompt, per engine, per check. The only tab you type raw results into.
Prompt Set
The questions you have decided you want to be the answer to. Ten to fifteen, tagged by intent.
Competitors
Who the engines name instead of you, counted automatically, with share of voice.
Legend & Dashboard
Every column explained, plus visibility rate, recommendation rate, splits by engine and by month, and the citation source breakdown. Nothing to type.
The method
How to run it yourself.
The whole thing is on this page. If you want to do it yourself, this is everything you need.
1
Build a prompt set of ten to fifteen questions
Not keywords. Questions, in the words a real buyer would use, typed the way people actually talk to an assistant. Cover four intents:
- Discovery. “Best SEO agency in Toronto.” The head term. Hardest to win.
- Problem-led. “My website gets traffic but no leads, what do I do?” Winnable early, because it rewards published thinking rather than authority.
- Comparison. “Should I hire an agency or do SEO in house?” Being cited here frames the decision before anyone builds a shortlist.
- Local. “Who does GEO in the GTA?” Small volume, high intent.
Then leave it alone. The most common failure is rewriting the prompt set every month, which gives you twelve months of anecdotes and no trend line. A mediocre prompt set tracked consistently beats a perfect one you keep editing.
2
Run every prompt in a clean session
No chat history. No personalisation. No logged-in account you have been using all year. Otherwise you are measuring your own account, not the model.
Run each prompt across every engine you care about. For most Canadian businesses that is ChatGPT, Perplexity, Google AI Overviews and Gemini. Log the result even when you are not mentioned: the zeroes are the baseline, and without them the improvement is unprovable.
Monthly is the right rhythm. Weekly measures noise and calls it progress.
3
Log the mention type, not just yes or no
This is the part that makes the tracker worth keeping. “Were we mentioned” is a binary that hides the entire story. The tracker splits it four ways:
RecommendedThe answer puts you forward. The only tier that behaves like a lead.
ListedYou are in the set of names. Real, but not the win.
Passing referenceNamed in a clause on the way to somewhere else.
Not mentionedThe baseline. Normal in month one. Not a verdict.
Going from Not mentioned to Listed and from Listed to Recommended are two different problems. Collapse them into one yes or no and you cannot tell which one you are solving. Log the position too: three or better, or it may as well not be there.
4
Read the citation source table
This is the table that decides your next quarter, and it is the one nobody builds. Every time an assistant names you, note what it cited. The tracker counts them and gives you a percentage. Then read it:
Mostly your own site
Keep publishing
The engines trust your content directly. Go deeper on the prompts where you are Listed but not Recommended.
Mostly third party
This is a PR brief
Your content is not the thing convincing them. No amount of on-page work moves this. The lever is digital PR: getting cited on the publications the assistants already reference.
Mostly Reddit
Go and read the thread
The engines are learning about you from strangers. Fine when the strangers are positive. An emergency when they are not.
No source at all
Pleasant, unrepeatable
The model is going from memory. You cannot defend it, and you cannot reproduce it.
That single table is the difference between “we should do more content” and knowing which lever actually moves the number.
5
Mine the competitor column
Every answer that names someone else is free competitive research, handed to you by the engine. Log the names. After three cycles the tracker gives you share of voice across your competitor set.
If the same three names keep surfacing, go and look at what the engines cite about them. That is your gap list, and it was assembled for you.
Calibration
Three ways people get this wrong.
Chasing the head term only
“Best agency in [city]” is the hardest prompt in your set and the one most saturated with directory listicles. The problem-led prompts are winnable this quarter.
Treating the engines as one number
Averaging across four engines hides the diagnosis. Invisible in AI Overviews and strong in Perplexity is a Google problem, not a GEO problem, and the fix is different.
Tracking without an action column
The Notes column exists so every check ends in a decision. Without it the monthly run becomes a ritual you resent, and the tracker quietly becomes a report nobody reads.
Questions
FAQ
Is the GEO tracker really free?
Yes, and there is no email form. Download it, edit it, rebrand it, use it with your own clients. We built it for our engagements and there is no reason to hide it. The people who will happily run this themselves were never going to hire us, and the people who open it and think “I am not doing this every month” are exactly who we are for.
How often should I check?
Monthly. AI answers move week to week for reasons unrelated to anything you did, so weekly checking measures noise. Same prompts, same engines, same rough date each month.
How many prompts should I track?
Ten to fifteen. Fewer prompts checked consistently beats fifty checked once. Cap your High priority prompts at five, because if everything is high priority then nothing is.
Does GEO replace SEO?
No, and be suspicious of anyone selling it that way. The assistants are largely reading the same web your SEO built. Strong technical foundations, real content and earned citations feed both. GEO is a measurement layer and a set of adjustments on top of that, not a replacement for it.
Will this work for a local business?
Yes, and local is often where the gap is widest, because assistants fall back on directories when local businesses have published nothing worth citing. That is an opening.
Can I use this in Google Sheets?
Yes. Use the Make a copy link at the top of this page. The dropdowns and colour coding carry across.
Or we can just run it
The tracker is the easy part.
The work is running it every month, reading the citation table honestly, and doing something about what it says. That is the engagement. We build the prompt set, run the checks across every engine, report the trend, and act on it, whether that means content, technical work, or digital PR to get you cited where the assistants are already looking.
