Ask an AI assistant to recommend an accountant, a software agency, or a project management tool, and it answers in seconds with three or four names. Those names did not get there by accident, and they did not get there by paying. They got there because, across everything the model has read and everything it retrieves at the moment you ask, they are the entities most strongly and most consistently associated with being the answer.
If your customers are asking assistants these questions, and increasingly they are, then how those names get chosen is now part of your marketing. Here is how it actually works, without the mysticism.
Two moments decide everything
An assistant's recommendation is shaped at two different times, and they reward different things.
Training time. The model has read a vast snapshot of the public web: articles, directories, documentation, forum threads, reviews. If your business appears in that material, clearly described and consistently named, the model carries an association between what you do and who you are. This is slow to build and slow to decay. You cannot buy it retroactively.
Answer time. For questions where freshness matters, most assistants now search the web mid-conversation, read a handful of pages, and compose an answer from them. This is the moment you can influence this quarter: the assistant is skimming a few sources and deciding, in effect, who is quotable.
Traditional SEO fought for a ranking on a page of ten blue links. This contest is harsher: the assistant reads the results, then names two or three businesses in a sentence. There is no page two. There is barely a page one.
What the machines are actually weighing
Across both moments, the same signals keep deciding who gets named:
Being an entity, not just a website. Models reason about things: businesses, products, people, places. A business that exists only as scattered pages is hard to reason about. One that is consistently named, described the same way everywhere, marked up with structured data, and referenced by third parties becomes a stable node the model can attach facts to. Entities get recommended; URLs get forgotten.
Being the cited source. When an assistant retrieves pages, it prefers ones that answer the question directly: a clear claim, evidence, no throat-clearing. Pages written as actual answers to actual questions get quoted. Pages written as brochures do not.
Third-party corroboration. A model is cautious about claims a business makes about itself. It is far more confident about claims others make: reviews, directory listings, press mentions, comparison articles. One good independent mention outweighs ten pages of self-description.
Machine readability. Assistants fetching your site under time pressure need to parse it fast. Clean HTML, real headings, structured data, fast responses, content visible without executing a pile of scripts. If a crawler gets a blank shell or a five-second wait, you were never in the running.
Consistency. If your name, services, and location are described differently across your site, your listings, and your profiles, the model's picture of you blurs. Blurry entities feel risky to recommend, and assistants are built to avoid risky answers.
What stops working
Just as important is what this filters out. Keyword-stuffed pages written for crawlers read as noise to a language model. Thin location-page sprawl reads as spam. Exaggerated claims without corroboration get discounted rather than repeated. A decade of tricks aimed at gaming a ranking algorithm is largely invisible to a system that reads like a person and cross-checks like a researcher.
The uncomfortable summary: you cannot trick your way in. The signals that persuade an assistant are mostly the signals of actually being a credible, well-documented business.
What to do about it, in order
- Decide what you want to be the answer for. One sentence: "When someone asks X, we should be named." Everything else serves that sentence.
- Say it plainly on your own site. A page that states what you do, for whom, with evidence, in prose a machine can lift straight into an answer.
- Fix the entity. Same name, same description, same facts everywhere: your site, your structured data, your directory listings, your profiles.
- Earn corroboration. Reviews, mentions, comparisons, and answers to real questions in places your buyers already look. This is slower, and it is the part that compounds.
- Stay technically legible. Fast pages, clean markup, structured data, content that renders without ceremony.
None of this is exotic. It is the credibility work most businesses postponed, made suddenly urgent because a machine now reads the whole record and picks winners out loud.
Where this is heading
Assistant recommendations will not replace search everywhere, but for "who should I use for X" questions they are already the first stop for a meaningful share of buyers, and that share only grows. The businesses being named today are building an advantage that compounds: recommendations beget citations, citations beget recommendations.
Our AI SEO work is exactly this discipline: making your business the entity assistants can find, trust, and name. If you want to know where you stand first, the Clarity Report checks the machine-readability layer, the structured data, speed, and the signals assistants depend on, free, in about a minute.