New Track whether ChatGPT recommends you
Your competitors are telling you what they are about to do.
They say it in pricing pages, changelogs, job adverts and the reviews their customers leave. You Compete Me reads all of it continuously, works out what the pattern means, and hands your team the work already drafted.
No card required Running in ten minutes Public sources only
You Compete Me App
HighNorthwind is preparing to move upmarket, and will leave their current base exposed.
Four signals in eleven days: SSO and audit logging shipped, two enterprise account executives hired in EMEA, the self-serve tier moved below the fold, and a 22% rise in reviews complaining about support response times.
- What would prove this wrong
- A self-serve pricing change, or a product update aimed at teams under 50 seats, within the next quarter.
- Evidence against it
- Their January webinar was explicitly aimed at small teams.
- Already drafted for you
- Objection handler for the “they’re going enterprise” call, and a win-back list of their SMB reviewers.
68% 4 signals · 3 quoted · 1 contradicting
AI search visibility
“Best vendor for mid-market ops” · 4 engines · this week
Northwind · support response time
Complaints per day, 30 days
+3.4σ against its own 14-day baseline
600
sources watched per account, continuously
4
answer engines probed — ChatGPT, Claude, Perplexity, Gemini
30‑day
window every reading is correlated across
100%
of conclusions carry a falsifier, enforced by schema
The problem
Alerts are not the hard part.
Plenty of tools will tell you a competitor changed a page. Nobody reads them for long, because a notification that something moved is not information — it is homework. The work is deciding what a run of changes adds up to, and that is exactly the part that gets skipped when nobody has an analyst.
| Approach | A spreadsheet | Page-change alerts | You Compete Me |
|---|---|---|---|
| What you get | Whatever someone remembered to paste in March | “This page changed” — forty times a week | A reading: what the pattern means and what it implies |
| Who does the thinking | You, on a Friday afternoon | You, after triaging the noise | The engine. You check its reasoning |
| Can you challenge it | Nothing to challenge | Nothing to challenge | Every reading names what would prove it wrong |
| What reaches the team | A link, if you remember | A notification nobody opens | The objection handler, already written, in Slack |
| Cost | Nothing, and it shows | Low, and it shows | From $149 a month |
What runs, continuously
Several kinds of watching. One conclusion.
You name the competitors. It finds the sources, reads them on a schedule, and only spends money on a model when something material actually moved.
The public record
Pricing pages, changelogs, blogs, status pages, job boards, press. Fetched politely, diffed against what was there before, and scored for materiality before anything is analysed — so a copyright year rolling over costs nothing.
Sentiment velocity
Not their rating. Which complaint is growing, how fast, and since when. Reviews are clustered by theme and tracked against each cluster’s own baseline, so you see a competitor failing while it is still happening.
AI search visibility
Buyers ask ChatGPT, Claude, Perplexity and Gemini who they should buy from. We ask the same questions on a schedule and track whether you are named, how you are described, and which sources get cited when you are not.
Work, already drafted
A reading becomes an artefact: the objection handler, the sales script, the campaign brief, the rewritten proposal clause. It lands in Slack, Teams or your CRM, assigned to whoever owns it, with a deadline.
Investigations on demand
Ask a specific question — why did they repackage, who are they hiring against — and it runs a multi-step research pass against a budget, returning what it found and what it could not establish.
Your weaknesses stay yours
Declare where you are weak and the engine stops recommending plays that attack a competitor on the same ground. Advice that would embarrass you in front of a buyer is worse than no advice at all.
The pipeline
From a changed page to a script in a rep’s inbox.
Every step is a background job. Nothing in the request path ever calls a model, which is why the dashboard is fast and the bill is small.
Collect
Fetch politely, honour robots.txt, use conditional requests so an unchanged page costs one 304.
Did it matter
Money and structure compared first, then normalised text, then near-duplicate detection. Below a materiality threshold it never reaches a model.
Enrich
A change becomes a typed signal with a verbatim quote. A job board cannot report a pricing change — the taxonomy forbids it.
Deduplicate
One announcement seen on the blog, the changelog, the press page and a tweet is one event, not a pattern.
Correlate
Thirty days of signals become a reading, with every contributing signal attributed — including the ones that disagree.
Act
The artefact gets written, routed to an owner by audience, given a deadline, and delivered where they already work.
Straight answers
What it will not do.
Worth saying plainly, because the category has a reputation for confident nonsense.
No private data, ever
Only what is public. No scraping behind logins, no bought data, no pretending to be a prospect to get a demo. Everything a reading rests on is something you could have found yourself — with a link to it.
No manufactured certainty
Confidence is calculated from the evidence — how many signals, how independent, how many carry a direct quote — and capped in code, not taken from the model. One signal cannot produce a confident conclusion.
No pattern every hour
“Nothing here” is a documented, expected outcome. A tool that finds a conspiracy every single hour is a tool that is inventing them, and you would stop trusting it within a fortnight.
Questions
The things people ask first.
How long until it is actually useful?
Minutes to set up, a few days to be interesting. Collection starts immediately, but a reading needs a pattern, and a pattern needs more than one signal. Sentiment clusters need a baseline before a spike means anything — reckon on a week for the velocity numbers to be worth reading.
Do I have to find all the sources myself?
No. Give it a competitor’s domain and it proposes the pages worth watching — pricing, changelog, blog, status, careers — and you approve or edit the list. You can add anything it missed, and remove anything that turns out to be noise.
What happens when it is wrong?
You will know, because every reading states what would prove it wrong. Reject the action it drafted and it asks you for a reason — that rejection reason is the only signal the system gets that it produced something that did not fit.
Is my competitor list visible to anyone else?
No. Isolation is enforced in the database itself with PostgreSQL row-level security, not by remembering to add a WHERE clause, and the application connects with a role that cannot bypass it. How that works.
Will you crawl aggressively and get my IP blocked?
No. robots.txt is honoured in full, including crawl-delay, one request at a time per host, and conditional requests so an unchanged page costs a single 304. The crawler identifies itself honestly in its user agent.
Can I get the data out?
Yes — export on every plan, and a public API from Growth upwards. It is your competitive picture; it should not be hostage to a subscription.
Fourteen days. Three competitors. The whole engine.
Name the competitors you actually worry about and leave it running. If after two weeks it has not told you something you did not already know, it has failed — and you have lost nothing but the ten minutes.