Guide
How to use KeenTag
Find out whether AI answers name you, fix one gap, then prove the fix worked with numbers that hold up. Three playbooks, each step linked to the page where you do it.
Who it’s for, and why they start
Founders and small SaaS teams
It starts when you ask ChatGPT for the best tool in your category and it names a rival, maybe one you have never heard of, but not you.
What they want: Who AI recommends instead of you, and one thing to do about it.
In-house marketers and SEOs
It starts when your boss or board asks “do we show up in ChatGPT?”, or traffic falls while rankings hold and you need to explain why.
What they want: A number that holds up when someone checks by hand, and proof that the work changed it.
Agencies
It starts when a client asks if they show up in AI answers, then types a question live in the meeting.
What they want: A defensible rate per client, a fix list to deliver, and a report link to send.
Why a rate with a range, not a rank. AI answers vary from one run to the next. SparkToro and Gumshoe found that the same question rarely returns the same list of brands twice (their study (opens in a new tab)). So KeenTag reports how often you are named across many answers, with the 95% range the true share likely sits in. How we measure
The core loop: see, fix, prove
See
KeenTag asks ChatGPT, Perplexity and Gemini your buyer questions many times and reports how often each answer names you, as a rate with its 95% range.
Fix
Each question where a rival is named and you are not becomes a fix: one specific change, in the order to do them.
Prove
After the change has had time to land, a re-test asks that question again. It counts as a real change only when the before and after ranges no longer overlap.
Playbook 1
Find out if AI recommends you, and who it names instead
Your first session. It ends with a rate per engine and a list of the rivals taking your spot. For: Everyone, first.
- Setup for step 1
Add your brand from your domain
Setup reads your site and suggests your name, rivals and buyer questions. Everything is editable.
Keep 3–5 rivals you actually lose deals to
Add any the suggestions missed and remove anything that isn’t a real alternative.
Keep about 10–15 questions a buyer would type
“Best X for Y”, “alternatives to Rival”, “X vs Y”. Delete anything that reads like your own homepage: those questions flatter you because nobody types them.
Keep the three engines separate
ChatGPT, Perplexity and Gemini are reported one by one. Don’t blend them in your head; they often disagree.
- Overview for step 5
Run the first check and read the rate with its range
A wide range means few answers so far, not a bad result. It narrows as more answers come in.
- Competitors for step 6
See who is taking your spot
Compare your rate with each rival’s. “Where rivals are named more” lists the clearest gaps first.
- Citations for step 7
See where the engines find your rivals
“Where to earn a mention” lists sites the answers cite often while naming you less than half the time. That is your off-site to-do list.
Playbook 2
Fix one gap and prove it worked
The loop that turns a diagnosis into a result you can defend. For: Everyone, every few weeks.
- Fix queue for step 1
Open the top fix
Each fix is one question where a rival is named and you are not, in the order to do them.
Do the change the fix describes
“Missing from a question”: publish or rewrite a page that answers that exact question near the top, in plain category words. “Thread to answer”: add a genuine answer to the thread the engines cite. “Source to earn”: earn a listing, review or mention on that site.
- Prompts for step 3
Leave the question’s wording alone
Same sentence every time. If the question drifts, the before and after no longer compare.
Give it time
Engines take days to weeks to pick up a change. Scheduled checks keep collecting answers meanwhile (Solo weekly, Pro daily, Agency daily).
- Prompts for step 5
Re-test the question
Open the question from the fix or from Prompts and press Re-test. It asks again now and records before and after.
- How a re-check decides for step 6
Read the verdict honestly
“Real lift”: the ranges no longer overlap, so call it proven. “No clear change yet”: they still overlap, so keep collecting and don’t claim a win. “Real drop”: look into what changed.
- Trends for step 7
Keep watching after a win
Answers drift as engines change their sources. Check the trend and re-test the same question again a few weeks later.
Playbook 3
Report to a client or boss
A number that survives someone typing a question live in the meeting. For: In-house teams and agencies.
- Overview for step 1
Run a check before the meeting
Fresh numbers first. Use Run check on the brand’s overview.
- Reports for step 2
Create a report for the period
Pick the period and create it. You get a share link that only shows that report.
Send the link or save a PDF
Copy the link, or open the report and use Save as PDF for a deck.
Walk through it in this order
The overall rate with its range, then each engine, then rivals, then fixes shipped and proven. Say “range”, not “rank”: AI answers vary too much from run to run for a rank to mean anything.
- Methodology for step 5
Invite the hand check
One question typed live is one sample. The rate is the share of many sampled answers, and its range already allows for that variation.
- Prompts for step 6
Show proof, not effort
Open a question whose re-test says “Real lift” and show that the ranges no longer overlap.
- Reports for step 7
Revoke the link when the work ends
Revoking a report link stops it working for everyone who has it.
The weekly routine
Once a week
A quick look: the rate and range per engine, and whether any new fix appeared. Nobody needs to watch this all day.
Overview (once a week)After each fix
Give it time, then re-test that question and read the verdict.
Fix queue (after each fix)Once a month
Compare yourself with each rival and pick the next gap to close.
Competitors (once a month)Once a quarter
Look at the trend by week and send a report to whoever needs it.
Trends (once a quarter)Tips
Should I accept every suggested question?
No. Review them. Keep the ones a buyer would type before choosing, and delete any that sound like your own marketing copy.
How many questions should I track?
About 10–15 that matter beats a long list. Start with the questions your sales calls already hear.
Can I reword a question to get a better result?
Don’t. A reworded question starts a new history. Add a new question instead and leave the old one as it is.
My range is wide. Is that bad?
It means few answers so far. More questions and more checks narrow it. A wide range is honest, not an error.
I typed the question myself and got a different answer. Is KeenTag wrong?
Probably not. AI answers vary from run to run, so one answer is one sample. The rate is the share of many sampled answers, with a range that allows for that variation.
How long until a fix shows up?
Days to weeks, depending on the engine and the change. Re-test too early and you will usually see “No clear change yet”.
A fix worked. Am I done?
Not quite. Engines change the sources they lean on, so a win can fade. Keep the question tracked and re-test it again a few weeks later.
Does KeenTag check how AI describes my brand?
No. It measures whether answers name you, not what they say about you. Each answer is kept in full, so you can read it.
Start with your own domain
Setup drafts your rivals and buyer questions from your site. You edit them before the first check runs.