Why ChatGPT Recommends Your Competitor Instead of You

ChatGPT recommends your competitor because it can identify them and it cannot identify you. Engines name businesses they can resolve into one consistent entity, cross-check against structured data that agrees with the visible page, and quote in a sentence or two. Three inputs: one unambiguous name, schema that matches the copy, passages written to be lifted. Miss one and you read as unclear, not as unworthy.

That paragraph is the whole post: answer first, mechanism second, short enough to quote whole. That shape is the format that gets pulled.

Nobody outside the labs knows the selection function, and anyone who claims they do is selling something. What you can do from the outside is run the questions yourself. Ask a hiring question tied to a place — who does humane bat removal in Grey County, best roofer near Mississauga for a flat commercial deck — then open the businesses it named and look for what they share. Three things repeat.

• It names businesses it can pin down. One spelling, one description, one consistent set of profile links resolves cleanly. A company that is Smith and Sons on the site, Smith & Sons HVAC Ltd. on one profile and Smith Heating on another is three faint signals, not one strong one.
• It prefers pages whose structure and prose say the same thing. When the markup declares a plumbing company serving Peel Region and the page reads like a plumbing company serving Peel Region, the two agree. Agreement is cheap confidence.
• It quotes what is quotable. A crisp forty-word answer under a question-shaped heading gets lifted. A nine-hundred-word essay with the answer implied in paragraph six does not.

Notice what is not on that list: tenure, workmanship, ad spend. An engine cannot taste your work. It can only read what you published. Your competitor is not better. Your competitor is more legible.

Three things, none of which needs a tool licence. One: whether your business is described identically everywhere it appears. Two: whether your structured data agrees with the words a visitor reads. Three: whether any single passage could be quoted whole, without the sentences around it. View-source and a search box cover all three, and most of what actually moves in AI SEO lives inside them.

Machines assemble your business from mentions, then decide whether those mentions describe the same thing. Every variation costs you — including the abandoned profile still advertising a service you dropped in 2019.

The check: paste your business name and one-line description from every place you are described — footer, directories, social profiles, association listings — into one document, verbatim. If that column does not read like one company, the machines are not seeing one company either.

The fix is boring and it works. One name. One description sentence. Both propagated everywhere, with your profiles linked from your site and your site from your profiles, so the graph closes. It matters most for businesses with no storefront, which is why service-area SEO without an address leans so hard on consistency.

Structured data is the machine-readable version of your page. Its value is not that it exists — plenty of sites emit schema nobody should trust. Its value is that it agrees with what a human reads. Markup declaring five services while the copy sells two contradicts itself, and contradiction is the condition under which an engine picks somebody else.

The check: right-click, View Source, search for application/ld+json, read what is inside. Three questions.

• Does the name in the markup match the name in the header?
• Do the services and the area served match what the page actually says?
• Is the business declared once, or several times in slightly different words?

Finding nothing at all is a finding too: every assistant reading your site is inferring your business from prose alone. The honest answer to which types you need is a short list, not the forty the plugin marketplaces sell.

Assistants answer in short spans. They need a passage they can take whole, not stitch together from four paragraphs. Write for that:

1. Make the heading the question a customer would actually type or say.
2. Answer it directly in the next forty to seventy words, as one self-contained block.
3. Then expand — detail, exceptions, cost drivers, local specifics.

Self-contained is the discipline: no pronoun pointing back at the heading, no wind-up before the answer. If the block would still make sense pasted into a text message with no context, it can be quoted.

Usually because the page never says. Below is a composite — not one real site — of the profile that gets skipped: a home-service business with a real reputation offline and a website a machine cannot read. Each row is something an engine looks for before naming you, and fails to find.

None of those rows is about content volume. The site is not too thin. It never states plainly who it is, where it works and what it sells.

We build to this standard on the brands we build, and you do not have to take our word for any of it. Open them and view source. Two of the three ship perfect Lighthouse scores — run it yourself.

• thecritterguy.ca — a twenty-year wildlife-removal company migrated off WordPress onto a static build. View source: one Organization entity, declared once, one name, the longer trading name declared as an alternate rather than left floating as a contradiction, and the social profiles declared in the markup so the graph closes.
• batsandwildlife.ca — built from zero, schema-first from day one. Its markup declares the towns it serves as named places, each pointing at its own page: a 37-town service-area architecture, so a question about one town has one page to match.
• cottagehero.ca — a property-care brand built to perfect Lighthouse scores from launch, no plugins, no database.

Static architecture is not a style preference here. A page with no database and no plugin layer renders identical markup to every reader — human, Googlebot, GPTBot — with nothing injected between source and answer.

One more piece of portfolio work we will describe but never name: a business rescued from a roughly 1,200-page spam infestation — cleaned, rebuilt, re-ranked, moved onto an architecture with no database and no plugins left to reinfect. We do not name clients in sensitive situations, and we never will. If an agency will publish that story with a logo on it, ask what they would publish about you. Method is shareable. Identity is not.

No, and you should decline anyone who says otherwise. There is no submission form, no inclusion fee, no lever to pull. You control the inputs — unambiguous entity, agreeing schema, citable passages, an architecture that renders the same markup to everyone. The engine decides the output. The rest of the limits, in the open:

• Outputs change week to week. The same question, same account, seven days apart can name a different set of businesses. Anyone showing a screenshot as proof of a permanent position is showing you a screenshot.
• Date your observations. Everything above is what we saw running hiring-intent questions in July 2026 — and the point is that you can run the same questions today and see for yourself. Write the date beside your own checks, then re-run the same wording in thirty days. An undated observation about AI search is worth roughly nothing six months later.
• It is not a separate discipline. Almost every input above is classic technical SEO done properly. Clean entities, correct structured data and answerable pages were good ideas when Google was the only judge. One investment, two arenas.

We build sites and rank businesses, and the entity work above is part of every build, not an upsell. Cost depends on real drivers: how many pages exist today, how many towns you serve, whether the site has to come off a platform first, how far apart your profiles have drifted. Tell us the questions you want to be the answer to and we will tell you what it takes. Tell us what you want to rank for — one form, and a real reply from the people who will do the work.