What AI Assistants Look For When They Recommend an Agency

It sees what its search systems can reach. Each company documents its own crawlers, and the documentation draws the same line: one crawler for search, one for model training, and one for pages a user asks it to open.

Table · scrolls sideways

Two things follow. First, an agency that blocks a search crawler, often by accident through a firewall or bot-protection setting, has removed itself from that assistant's search answers, however good its work is. Second, blocking a training crawler is a separate decision. OpenAI says disallowing GPTBot "indicates a site's content should not be used in training", and Anthropic describes ClaudeBot the same way. Neither is the switch for search visibility.

Because "best" is not a fact an assistant can look up. No neutral register of the best agency in Kitchener-Waterloo, Ontario or Canada exists. What the assistant can find is a set of pages that answer the question in parts: who offers this service, where they are, what it costs, what they have done, and what others say about them. A firm that answers those parts plainly is easier to describe accurately than one whose homepage says only "crafting digital excellence".

Google's guidance is the clearest public statement of what that means in practice. It says "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary", and that "all existing SEO fundamentals continue to be worthwhile" (AI features and your website). Its people-first content page says its systems "prioritize helpful, reliable information that's created to benefit people", and that of the qualities it looks for, "trust is most important" (Creating helpful content). OpenAI, Anthropic and Perplexity do not publish a ranking formula for which businesses their assistants name, so anyone who claims to know one is guessing.

Note

Citable page: a public page that a search crawler is allowed to reach, that states a checkable fact in plain words near the top (who you are, where, what you do, what it costs), and that says the same thing as your other pages and profiles. An assistant can quote it and attribute it without having to guess.

The last part is entity consistency. Google says Organization structured data helps it "disambiguate your organization in search results", and recommends properties such as description, address and sameAs. The markup only helps if it agrees with the visible page and with your directory listings. If your site says Kitchener, your Google Business Profile says Waterloo and an old directory says Toronto, a system reading all three has to pick one.

Table · scrolls sideways

Third-party mentions sit firmly in the right-hand column. A client writing about your project, a local publication covering you or a genuine review is evidence an assistant can weigh precisely because you did not write it. That is also why it cannot be bought honestly, and why a vendor promising "AI placements" deserves the same scepticism as one promising first place on Google.

Our own site is the example we can describe without guessing. As of 17 October 2026, its robots.txt allows OAI-SearchBot, Claude-SearchBot and PerplexityBot by name; it serves an llms.txt generated from the same route list as the sitemap; and every page carries the same Organization node, built from the same record as the visible pages. Our prices are published on the pricing page in Canadian dollars plus HST. We publish no review counts or awards, and no client figure without the client's permission, because an assistant repeating an invented figure is worse than one repeating nothing.

None of that guarantees a mention. It makes the facts easy to find and hard to misstate.

For a sense of how rare even basic machine-readable facts are, our Ontario Practice Website Index measured Ontario practice homepages (edition 2, October 2026, n = 187): a convenience sample of law, clinic and accounting practices, not all Ontario businesses. 150 of the 187 carried some schema.org markup, 53 served an llms.txt, and 8 showed prices on the homepage. Price is one of the first things a buyer asks an assistant, and most of those sites gave it nothing to quote.

Treat it as a list of leads, not a verdict. Then check each name the way you would check a directory listing:

1. Open the firm's own website. Confirm it exists, names a city, and offers the service you asked about.
2. Look for a published price or a stated pricing basis. If the assistant quoted a price, find it on the firm's site.
3. Read one case study end to end. It should show a problem, a process and what shipped, with the client named or the work labelled as anonymised or concept.
4. Ask the assistant for its sources. If it cannot point to a page, treat the claim as unverified.
5. Put the same questions to every firm. Our questions to ask before hiring a web design agency is a sheet you can send as it is.