Here is a situation that comes up constantly.
A business repairs its site. Schema is valid. Content answers real questions. Crawlers are welcome. Everything on the checklist is done.
Then someone asks ChatGPT for a recommendation in that category, and it names a competitor with an obviously worse website.
The instinct is that the work failed. Usually the work was fine and incomplete, because the missing layer was never on the website.
Self-description is cheap and engines know it
Everything a business says about itself on its own site is a claim. Claims are free. Anyone can write that they are the leading provider of anything.
Generative engines are built with that in mind. What raises confidence is not the strength of your claim, it is the number of independent sources that describe you the same way. Your homepage saying you are an Arizona cinematic production studio is one data point. Nine unrelated sources describing you in those terms is a different kind of evidence.
This is the corroboration layer, and it is where most businesses have done nothing at all.
What an engine is actually doing
When an assistant answers a recommendation question, it is not ranking pages the way a search engine does. It is assembling a description of entities it is confident about, and confidence is the constraint.
An entity it can describe confidently gets named. An entity it can only describe tentatively, or where sources disagree, gets omitted in favour of one it can state plainly. Omission is not a penalty. It is the safest output when the evidence is thin.
Which means the competitor with the worse site is not beating you on quality. They are easier to describe.
The surfaces that corroborate
Corroboration comes from sources that are not you, describing you, in terms that agree.
- Category-relevant directories and platforms. Not every directory. The ones that are actually used in your category, filled out completely and consistently.
- Industry associations and certifications. Membership and credential listings are independent, structured, and treated as credible.
- Supplier, partner, and vendor pages. If you appear on a partner's site as a named vendor, that is a third party describing your role.
- Local and trade press. Coverage carries both the description and the credibility of the publication.
- Interviews, podcasts, and guest contributions. These attach a named person to a named business and a described specialty, which strengthens the operator and the entity at once.
- Review platforms. Customer language describing what you did, in words buyers actually use.
Unlinked mentions count
This is where entity work departs from classic link building.
In link building the hyperlink was the asset. For entity confidence, the described association is the asset. A page that names the business alongside its city and its service contributes to understanding whether or not anyone linked anything.
That reframes the goal. You are not chasing link equity. You are trying to make the true description of your business appear, consistently, in more independent places.
Why buying it backfires
The obvious shortcut is to purchase placements at volume. It does not work, and it is worth being precise about why.
Paid link networks produce mentions that cluster: similar page templates, similar anchor phrasing, similar timing, similar neighborhoods of the web. Those patterns are exactly what ranking and retrieval systems have spent two decades learning to identify and discount. The spend buys a footprint that reads as manufactured, and manufactured footprints are treated as evidence of manipulation rather than evidence of the business.
The slow version works because it cannot be faked cheaply. That is the entire mechanism.
The timeline is the honest part
Technical repair moves quickly. Fix crawlability and indexing and pages can move within weeks. Fix the Google Business Profile and results often show inside one to two months.
Corroboration does not behave that way. It accumulates. Its effect on how AI answer engines describe and recommend a business generally shows across three to six months, because those systems re-crawl, re-evaluate, and update on schedules nobody outside them controls.
Any vendor promising AI citations next month is either misunderstanding the mechanism or misrepresenting it.
Citation building on the directories, profiles, and platforms that corroborate the entity is part of the Expand and Compound monthly plans at UM Media, which follow the SEO / AEO / GEO Foundation. Structure only. No ranking or citation guarantees.