A daily brief on what your competitors actually changed
Each morning it compares today's prices and plans against yesterday's, works out what genuinely moved, and writes a short brief on what it signals. The quiet days produce nothing, which is the point.
Tested 26 Aug 2026 on 4 sample competitors: 2 real changes reported, 2 filtered out.
How it works
Five steps, left to right. Click the picture to see any step larger.
Today's prices, next to yesterday's
It runs at 7am, or whenever you ask it to, holding two records per competitor: yesterday's pricing page and today's.
A plain comparison finds the differences
Not the AI. Simple matching, value by value, so a change is either there or it isn't.
- Northstar, Growth
- was 89 per month
- Now
- 109 per month
Page noise is dropped here
A footer timestamp or a session id changes on every visit, so those fields are left off the watch list.
- Cobalt Insights
- only the footer changed
- Result
- not reported
The AI (Claude) explains what it signals
It sees only the differences already found, never a whole page, and says whether the move needs an answer this week.
One brief, biggest news first
Sorted so the expensive news is read first. On a day when nothing moved, it says so in one line.
- Brief header
- 4 monitored · 2 changed · 2 unchanged
How it's built
Three decisions that decide whether a brief like this is worth reading after a week.
The comparison runs before the AI, not after
The obvious build hands the AI both pages and asks what changed. That pays for every unchanged word, and a model asked to find a difference will offer one whether or not it exists. Here plain code finds the change — exactly, and for free — and the AI only interprets what was already found. It cannot report a price move it never saw.
Chatty fields are deliberately not watched
A pricing page footer usually carries a timestamp and a session id that change on every visit. Watch those and every competitor looks like they changed something every day, and the brief is worthless within a week. Two of the four test competitors exist to prove the filter: one changed only footer text, one changed nothing at all. Neither reached the AI or the brief.
A number the data never stated is an invented fact
After the first live run the brief wrote "$89" — but the source records carry no currency at all. The AI had added a plausible detail that nobody gave it. Its instructions now forbid currency symbols outright, so it writes "89 per month". A small thing, and exactly the kind of thing that makes a reader stop trusting the rest.
What was verified
Run end to end on 26 August 2026, on four sample competitors chosen so the comparison had to tell them apart: one raised a price, one renamed a plan and added two others, one changed only its footer, and one changed nothing.
The items were counted at every step rather than assumed from a green tick: 4 competitors in, 4 compared, 2 through the filter, 2 brought back together, 1 brief out. 11 steps, 11 of 11 connections wired, no errors.
Not yet verified: it reads built-in sample records, not live pricing pages. Reading a real page means adding a fetch step and somewhere to keep yesterday's copy — the comparison, the filtering and the brief would not change.
From the test run
HIGH — Northstar Analytics. Growth moved from 89 to 109 per month, a 22 per cent rise, with no change to seats or trial length.
Two competitors produced nothing at all, and that was the correct result for both. A brief that reports something every day is a brief nobody reads.
