Route each finding before you fix it
The same symptom, missing from an answer, has different causes and different fixes. The diagnostic from chapter 3 decides which:
| Finding | Cause | Work |
|---|---|---|
| Described vaguely or wrongly when asked directly | The engine does not understand the claim | Your own pages |
| Understood, but not named in category answers | Not enough independent support for the claim | Third-party sources |
| A wrong fact traced to someone else's page | The source is out of date or mistaken | Correct it at the source |
| A recurring objection | A real concern, stated without context | Both: your pages and the sources |
Your own pages: make the claim easy to lift
Work through three questions for each claim:
- Is there a page whose main job is this claim? If not, that page is the first piece of work.
- Which pages already get cited or fetched? Those are the places where stating the claim reaches an engine soonest.
- Is the claim stated plainly, near the top, in a sentence that still makes sense quoted on its own?
Then the structure: headings phrased as the questions buyers ask, a direct answer under each, a short FAQ where it fits, structured data that matches the visible page, a date and a named author. A model is more likely to quote a sentence it can lift cleanly than a point buried in paragraph four.
Answer objections without arguing
An objection that keeps appearing is usually true for someone. Say who: the situation in which the concern is fair, the situation in which it is not, and the evidence for the second. A page that pretends the concern does not exist gives an engine nothing to weigh against the reviews that raise it.
Other people's pages: change the consensus
Before contacting anyone, write a one-page reference: the claims, the evidence for each, the facts that are commonly wrong, and the correct versions. It keeps every request consistent, whoever sends it.
Then work outwards in this order:
- Fix the mentions you already have. Out-of-date pricing, old product names and wrong descriptions on pages that already name you are the cheapest wins. Start with partners, integrations and customers, where you have a relationship.
- Get onto the pages that are already cited without you. Comparison articles, lists and directories from your baseline's source list. Earn the place with evidence and a clear reason to include you.
- Create new sources. Original data others want to reference, practitioner writing on credible sites, and genuine participation in the communities your buyers use.
Correcting what an engine gets wrong
- Record the exact prompt, the engine and the full answer, so the error can be reproduced.
- Find where it came from: look at the sources cited, and search for the wrong claim itself.
- Fix your own pages first, so the correct version is clearly stated somewhere authoritative.
- Ask the owners of third-party sources to correct theirs, with your reference page attached.
- Re-run the prompt on a schedule rather than once, because answers vary from run to run.
Check that it moved
Re-run the same prompts, unchanged, and look at the same things you measured in the baseline: how often you are named, which sources are cited, how the answer describes you, and whether the facts are right.
Give it time. Engines that search the web can reflect a change within weeks; what a model learned in training changes on a much slower cycle. Read the trend over a few weeks, never a single day.
Key takeaways
- Not understood means your own pages; understood but not recommended means other people's.
- Give each claim a page, state it plainly near the top, and structure it to be quoted.
- Answer objections with context and evidence rather than denial.
- Fix existing mentions first, then get onto pages already cited, then create new sources.
- Re-run the same prompts and read the trend over weeks.