When you retract a claim, the copy is the easy part
The pages you think of are the pages you fix. The claim lives on in the places that generate output nobody reads.
A site decided a phrase it had been using overstated what it could support, and changed it. Homepage updated, product pages updated, the change reviewed and signed off.
Months later the original phrase was still on five long-form articles and, worse, in the description of a product feed being submitted to a shopping engine. The retraction had been applied to everything a person would naturally open, and to nothing else.
Where claims hide
- Long-form articles. Usually a closing paragraph about the company, written once and never revisited.
- Generated feeds. Product and merchant feeds are built from templates in code. Nobody opens them, and they publish continuously.
- Structured data. Descriptions inside schema markup are invisible on the page and read by machines, including the ones writing AI answers.
- Metadata. Page descriptions, social preview tags, keyword tags left over from an older template.
- Transactional email. Order confirmations and dispatch notices tend to be the last thing anyone audits.
Do it as a search, not a review
Grep the entire repository for the phrase and its obvious variants, then work the list. Do not rely on remembering where it was used.
grep -rin "phrase\|variant\|near-synonym" --include="*.js" --include="*.html" .Expect two categories in the results, and treat them differently. Some hits are the claim being made. Others are content legitimately discussing the concept. In the case above, an article explaining what the term means in general and warning readers to be sceptical of it was entirely correct and should not have been touched.
Check whether your own published advice contradicts your own marketing. That site had an article telling readers to distrust suppliers who make exactly the claim it was still making in five other places. That is worse than the original overstatement, and it is only findable by reading both.
Verify from outside
Confirm against the live site rather than the source, because generated output can lag a deploy and templates can be cached. Fetch each affected page and the feed, and count occurrences. Zero is the only acceptable answer.
The general rule
Any statement worth retracting was probably worth reusing, which means it was probably templated, and templates end up in outputs that no human opens. When you change a claim, search for it rather than reviewing for it.
Need help with any of this?
These notes are free and always will be. If you would rather someone just set it up, or you are stuck on something similar, get in touch at hello@opsira.io.