AI Search Tracker — Will ChatGPT Mention You? GEO Visibility for Products
Add your products (name + URL) — we test 25 buyer prompts across ChatGPT, Perplexity, Gemini & Copilot weekly. Get mention rate, position, sentiment, citation — and what to fix when competitors win.
Purchasing answers are
moving to AI.
“Best CRM for agencies,” “top running shoes under $150” — the answer is now a synthesized recommendation that names specific products. If yours isn't named, it doesn't exist in that moment of intent.
The challenge is measurable
AI Search Tracker turns "are we visible to AI?" into a 0-100 score built from mention rate, position, and sentiment across the whole buying funnel — so it degrades exactly when the AI stops recommending you.
Weekly, not monthly
AI answers shift as models update and content recirculates. A monthly check shows what happened; a weekly check shows what changed — before it costs you a quarter of pipeline.
The GEO playbook
Understand the mechanics in generative engine optimization, then read how the visibility score is built and how to read the gap reports in order of impact.
Track. Score.
Fix what competitors win.
01
Add products
Name + URL per product. Auto-seeds 25 prompts (awareness / consideration / decision) and runs a baseline scan so you know where you start.
02
Track weekly
Cron tests prompts × engines, recording mention, position, sentiment, citation, and answer type for every single answer.
03
Act
Scorecard + action loop: schema, llms.txt, PR when not mentioned or outranked — see exactly which fix to run first.
Every signal that decides
whether AI recommends you.
Mention rate.
What share of your tested prompts name your product at all. This is the closest thing to an "AI share of voice" and the headline number most teams care about.
Position.
When you are named, are you first, in the middle of the list, or a footnote? Position tracks your status within the recommendation, not just your presence.
Sentiment.
The AI isn't just listing you — it's arguing about you. Positive ("top choice for…"), neutral ("one option is…"), or negative ("lacks…") changes what the mention is worth.
Citation.
Whether your own site or profiles are linked as a source. Cited brands reinforce entity and authority the way backlinks do in classic SEO. The action loop: weak prompts map to specific fixes — llms.txt entries, FAQPage schema, comparison and spec content, and digital PR.
Example next actions when a prompt is lost
- • Not mentioned, no citation → add
llms.txt+ FAQPage schema for the exact category - • Mentioned but negative sentiment → strengthen trust signals: backing, comparisons, testimonials
- • Mentioned 3rd but competitor is 1st → publish a direct comparison page + get a third-party citation
- • Cited wrong or stale details → enforce consistent NAP/entity data across site and listings
For the reasoning behind each fix, see reading the gap reports in order of impact and entity SEO for AI search.
One scan infra,
two views.
25 prompts per product
Auto-seeded across awareness, consideration, and decision — the full buying funnel.
Multiple engines
ChatGPT, Perplexity, Gemini, and Copilot tested weekly on the same prompts.
Mention rate
What share of your prompts name your product at all — your AI share of voice.
Position within answers
First, middle, footnote, or absent — where you sit inside the recommendation.
Sentiment
Positive, neutral, or negative framing of each mention, counted run over run.
Citation tracking
Whether your site or profiles are linked as sources — AI's version of backlinks.
0-100 visibility score
Mention rate + position + sentiment rolled into one number you can move.
Action loop
Lost prompts map to concrete fixes — llms.txt, FAQPage schema, comparison content, PR.
Per-user isolated
Products, prompt sets, and history stored per account behind RLS — agency-safe.
Rank trackers show keywords.
This shows whether you're being recommended. Traditional rank tracking answers “where does my page show up for this keyword?” — still useful, but it doesn't tell you whether ChatGPT, Perplexity, Gemini, or Copilot will name your product. Those engines don't index page rankings; they synthesize recommendations from entity data, citations, and content they can parse. See the honest overlap in AI visibility vs Google rankings.
Apples-to-apples across engines
The scorecard is intentionally the same shape across engines, so “mention rate on ChatGPT” versus “mention rate on Perplexity” is a real comparison. If a product shows up on Gemini but not ChatGPT, you're seeing a genuine distribution difference — exactly the measurement most teams are missing today.
How it fits your existing analytics
Reuses lib/analytics/runAnalyticsScan.ts + scan_responses / metrics_history — now product-centric, not business-centric. Separate tables ai_tracker_products etc., per-user RLS, isolated billing. The same engine powers the RankBot robot (auto-fix + PR from weak prompts) and the weekly AI rank report: a product scorecard here, a full marketing-site audit over here. For teams just starting, read what a free AI visibility scan reveals before paying for anything.
Product AI tracking,
answered.
What exactly is being tested?
Each product gets 25 buyer prompts spanning awareness, consideration, and decision. Every weekly run queries those prompts across engines and records mention, position, sentiment, and citation per answer.
Is this the same as Google rank tracking?
No. Rank tracking checks keyword positions on search pages; this checks whether AI answer engines name and recommend your product. The two overlap, but AI recommendations are synthesized from entity and citation signals, not page-one placement.
How does the 0-100 visibility score work?
It combines mention rate, position within answers, and sentiment into one number. Brand new tracking starts with a baseline scan so you can watch the score move as fixes land.
What do I do when a competitor wins a prompt?
The dashboard maps the lost prompt to a fix: schema, llms.txt, comparison content, or digital PR. RankBot can take the same signal and open a pull request with the fix automatically.
Which engines are supported?
ChatGPT, Perplexity, Gemini, and Copilot are tested by default, with the pipeline built to add engines as they become measurable.
Do I need a login?
Yes — products, prompt sets, and history are stored per user behind RLS, so an agency can track each client's products with full isolation.
If AI answers the question,
be the answer.
Test your products against 25 buyer prompts across the engines your customers ask — weekly, scored, and fixing the gaps.