Home / Tracking platform
Case 3 · Deployed · Our own product
Can you track what AI says about your firm month to month? This is how we do it.
This is the tracking platform behind our AI visibility audits for Ontario B2B service firms. It asks a business's buyer questions in ChatGPT, Perplexity, Gemini and Claude on a schedule, and records who gets named and which sources each answer cites.
Why isn't one check enough?
Because every engine varies from run to run. In our own audit, Perplexity named our agency in 2 answers in October and 0 in November, on the same questions. The overall drop, from 8 of 75 to 5 of 75, was inside that noise.
One check tells you where you stood that day. To know whether a fix worked, you need the same questions asked again and again, with every answer and citation kept. Typing questions into ChatGPT by hand doesn't give you that, so we built something that does.
What does it do, step by step?
-
Starts from a website
Claude reads the business's site and drafts 50 buyer questions and a list of competitors.
-
Asks the engines
Each question goes to ChatGPT, Perplexity, Gemini and Claude.
-
Records every answer
Who was named, and every source each answer cited.
-
Logs AI crawler visits
When AI crawlers fetch the business's own site.
-
Finds openings in search results
Pages and listings the business could appear on.
-
Builds an action plan
Each item graded by its likely impact.
-
Weekly scans when scheduled
The same questions run again, so the next result sits beside the last one.
How much has it recorded?
13,206
citations: the sources the engines pointed to in their answers.
468
prompt runs: a question asked in an engine, with the answer kept.
251
prompts: buyer questions on file.
43
tenants. A tenant is one business set up for tracking.
| Record | Count |
|---|---|
| AI crawler visits logged | 360 |
| Opportunities found in search results | 67 |
| Competitors on file | 31 |
Row counts from the live database, counts as catalogued on 6 October 2026. They're records in the system, not a count of paying clients.
What is it built with?
- Supabase Postgres: 40 tables, with row-level security on every table.
- Supabase Edge Functions for scans, reports and onboarding.
- pg_cron for scheduled weekly scans.
- n8n workflows, including onboarding a new business.
- DataForSEO for engine answers and search results.
- Claude (Anthropic) to read a website and draft the first questions.
What can't it see?
- Copilot. No API we use returns Copilot answers, so in an audit and in the monthly re-check we check Copilot by hand on Bing and Bing AI Mode, signed out. How we do that.
- Signed-in answers. It can't see what a buyer sees inside their own account, with their history and work data.
- Every possible wording. Its questions are a sample of how buyers ask, not all of them. That's why we write audit questions with you, from your real leads' words.
- The answers themselves. It records what engines say. It doesn't change it. The fixes do that, if anything does.
What does this mean for my firm?
Your audit's questions, up to 30 of them, become a fixed set. If you take the optional $99/month re-check, those same questions go to the same seven engines as the audit, including Copilot (Bing AI Mode), checked again each month and compared with your audit, with a short note on what moved and what didn't.
That's how you find out whether the Bing listing, the price page or the Clutch profile made any difference, instead of guessing.
Start with a baseline
You can't track what you haven't measured once.
The $499 audit gives you the baseline: your buyer questions, who gets named instead of you, and the fixes to make first. Book a free 30-minute check to see whether it's worth it for your firm.