The problem nobody budgets for
Here's a pattern we see constantly. A company changes its price. The pricing page gets updated. The FAQ doesn't. A blog post from last year still has the old number. And a third-party roundup that ranks above all of them has a number from two years ago.
Now someone asks a chatbot what it costs. The model sees four different answers. It picks one, or it hedges, or it cites the roundup. In every case, you lost the answer to your own question.
This isn't a writing problem. It's a data problem that shows up in writing.
Entity truth
Search and answer engines think in entities: your company, your products, your people, your competitors. Each entity has attributes. Price, founding date, what the product does, who it's for.
Entity truth means every attribute has exactly one canonical value, and every surface you control says the same thing.
The claim library
We keep a claim library for every portfolio company. It's boring on purpose. Each row is one claim:
| Field | Example |
|---|---|
| Claim | Monthly price, starter plan |
| Value | $99 per month |
| Effective date | 2026-10-01 |
| Source of truth | Billing system, plan ID |
| Owner | finance-agent, approved by founder |
| Where it appears | Pricing page, FAQ, schema, 3 briefs |
The rule
Every page, every schema block and every content brief must match the library. If it doesn't, it doesn't ship.
That's claim lock. When a claim changes, it changes in the library first, and everything downstream gets updated in the same pass. No orphaned numbers.
Citation hygiene
You can't edit the whole internet, but you can make your own domain the obvious source. Publish the canonical fact on a page built to be cited. Ask the roundups that rank to update their numbers. Make sure your structured data carries the same value as your copy.
For reference sources like Wikipedia, the rule is simple: correct errors through the proper channels, with disclosure. Never make undisclosed edits. It's wrong, and it backfires.
Where agents help
This is perfect agent work. It's repetitive, it's rule-based, and humans forget to do it. Our data-engineer agent crawls every owned page weekly, diffs every claim against the library, and flags drift. Changes to the library itself go to a human. Agents enforce the truth. People decide what it is.
Boundary conditions
Some lines we don't cross, no matter what a test says. No rating schema without verified ratings. No invented statistics. No claims a regulator would object to. Trust compounds slowly and breaks fast.
Building in a legacy industry?
We bring the agent crew and the distribution. You bring what you know about the industry.