- 01HabiticaTurns goals into a game
- 02BeeminderMoney on the line
- 03NotionBuild your own tracker
Measure AI presence
AnswerFit tests how AI models describe and recommend your product on the questions your customers ask. Get an evidence-backed report showing where you are missing or misread, which changes are worth testing and how answers compare after a re-test.
Watch a runThe run
ANNET / ILLUSTRATIVE RUNAn illustrative four-act run for Annet, an Anphase goal and accountability product at annet.ai. Ask: four models answer the same question about staying accountable to a big goal and none name Annet. Measure: Annet's report recorded on 3 Sept 2026 finds Annet named in 22 of 24 answers asked by name and 12 of 114 answers when customers ask. Across those 114 customer answers, Habitica is mentioned in 40, Beeminder in 38, Notion in 36 and Annet in 12. Change: category copy, SoftwareApplication JSON-LD and llms.txt are added, and the text that carries the argument is server-rendered. Re-test: Annet is named first by ChatGPT and Gemini, second by Claude and not named by Grok, for three of four sample answers, with an illustrative Fit Lift of +0.21, interval +0.12 to +0.30, and negative controls held. The baseline figures in Measure are recorded; the answers, the changes and the re-test are illustrative, not a measured lift.
Examples
Also mentioned
- Insight Timer
- Calm
- Headspace
Priority to test
Selfime's site publishes no llms.txt file.
Show history and next stepsHide details
What a model found, crawled 8 Oct 2026
- no structured data
- no llms.txt
- 35% readable text
Changes so far
- 29 Aug: 8% readable text, no structured data, no llms.txt
- 8 Oct: readable text up to 35%; the other two gaps remain
- Prepared, not live: llms.txt and structured data, waiting for review
Next to try
- Add structured data that says what it is
- Put the key text in the page itself
Re-test after the top fix
Simulated to show what a re-test would report, and not a measured result.
Also mentioned
- Zotero
- Readwise Reader
- NotebookLM
Priority to test
No page on Poreover's site carries structured data saying what it is.
Show history and next stepsHide details
What a model found, crawled 8 Oct 2026
- 3% readable text
- no structured data
Changes so far
- 29 Aug to 3 Sept: little readable text, nothing for models to read
- 8 Oct: still one page at 3% readable text
- Prepared, not live: llms.txt, structured data and a clearer first line
Next to try
- Put the key text in the page itself
- Link the name to its official profiles
Re-test after the top fix
Simulated to show what a re-test would report, and not a measured result.
Also mentioned
- Strides
- Todoist
- Motion
Priority to test
Add schema.org sameAs on the entity's JSON-LD pointing at owned, unambiguous profiles.
Show history and next stepsHide details
What a model found, crawled 8 Oct 2026
- 5% readable text
- 2 pages empty without script
- no llms.txt
- structured data present
Changes so far
- 22 Aug: 4% readable text and no structured data
- 2 Sept: structured data added to the site
- Prepared, not live: llms.txt and links to its official profiles
Next to try
- Say in the first sentence what it is not
- Pair the name with a unique qualifier
Re-test after the top fix
Simulated to show what a re-test would report, and not a measured result.
Also mentioned
- 0x0.st
- transfer.sh
- webhook.site
Priority to test
No page on svc.nz's site carries structured data saying what it is.
Show history and next stepsHide details
What a model found, crawled 8 Oct 2026
- no structured data
- 38% readable text
Changes so far
- 29 Aug: 20% readable text, no structured data
- 8 Oct: readable text up to 38%; structured data still missing
Next to try
- Put the key text in the page itself
- Publish a page naming the specific capability and its mechanism
Re-test after the top fix
Simulated to show what a re-test would report, and not a measured result.
What your report gives you
Evidence for your next decision
Each assessment brings the findings, answer evidence and priorities together in a client report. You can read the summary, inspect the supporting detail and keep a complete record of the answers.
Public-site review
See which product facts are readable on your public pages, with missing or unreadable evidence called out.
Questions that matter
Separate answers that mention your name from answers to customer questions, so recognition cannot hide a recommendation gap.
Priorities to test
Each suggested change carries its supporting evidence and testing status, helping your team choose what to try first.
Comparable re-tests
Repeat the same questions after a change and see what improved, stayed unchanged or moved backwards, with uncertainty stated.
We assess your public site without editing it. Controlled tests help assess a proposed change; a live re-test is needed to learn whether published changes affected the answers people receive.
How it works
Each run is versioned so every claim can be retested against a comparable run.
Four readings, never one score
Does AI understand what you made?
We measure recognition, description accuracy, suitability and recommendation separately. Your report distinguishes a model recognising your name from it recommending your product for a customer's need.
Known by name
When someone asks about you directly, does the model recognise you?
Described correctly
When it does, is what it says true, or is it describing something else?
Suited to a use case
When someone describes a problem you solve, are you named?
Recommended
When someone asks what to choose, are you among the answers?
Your team keeps a record of each hypothesis, the evidence behind it and the result of the re-test. Findings that did not move are retained alongside improvements, so the next decision starts from the full record.
In private beta
Bring your product URL and the questions you want AI to answer accurately. We will use them to scope an assessment and agree which models, evidence and re-tests matter for your team.