How an AI answer picks which roofer to name.
The retrieval and attribution mechanism, described without invented statistics.
Retrieval, then attribution
A system answering a question about roofers in your city retrieves candidate pages, extracts passages, and attributes the result to sources. Google's AI Overviews draw on Google's own index. ChatGPT browsing and Perplexity fetch pages at query time using their own crawlers and, for some queries, third-party indexes. Being crawlable by more than Googlebot therefore matters in a way it did not five years ago.
What makes a company nameable
Five properties, none of which guarantee a mention.
Why consistency matters more here
When an answer appears without a click, the system has to decide who it is describing. Two addresses, three phone formats, and a business name that differs between the site and the directories give it reason to hesitate or to merge you with someone else. Entity clarity is unglamorous clerical work, and it is the part most roofing sites have never audited.
The role of third-party mentions
These systems do not only read your site. They read the directories, review platforms, supplier pages, association listings, and local news that mention you. A contractor with a thin site but consistent, substantive mentions across those sources can be named more readily than one with a good site and no external presence. Both halves are the work.
What you cannot control
Which sources a given system trusts, how it weights them, whether it cites anyone at all for a query, and how often that changes. All four move without notice. That is why nobody can guarantee citations, and why the only durable strategy is being genuinely the best available answer to specific questions rather than gaming a retrieval mechanism that will have changed by spring.
The questions where roofers have an opening
Specific, local, and largely unanswered ones. What a hail claim looks like in a particular state. How a tile repair differs from a shingle repair on a two-storey home. What a TPO specification should contain for a warehouse of a given age. Volume-led content strategies skip these because the search volume looks small. Answer engines do not care about volume. They need an answer to attribute, and for many of these there is not a good one published by anyone.
How to test it on your own market
Ask the systems. Run the queries your buyers use in AI Overviews, ChatGPT, and Perplexity, record which companies get named and which pages get cited, and repeat monthly. It is manual and imperfect, and it is considerably more honest than a dashboard metric nobody can define. It is also what the citation checks in an audit do, at more scale.
Make sure more than Googlebot can reach you
Four checks, all quick.
Whether to block them instead
That is a legitimate choice for a publisher selling content. For a roofing contractor whose goal is to be recommended, blocking the systems that do the recommending is the wrong end of the trade. The decision is different if you host substantial original research you licence, which almost no contractor does.
What a good answer about your company reads like
Ask a system who the best roofers in your city are and read what it says about the ones it names. It will typically state what they do, where, how long they have operated, what they are certified in, and something specific about their work. Every one of those facts came from a page or a citation. If your site does not state them plainly, there is nothing to assemble that paragraph from.
Related guides.
Two acronyms for the same shift: being the source an answer engine quotes rather than a link it lists.
Nine checks, each verifiable before launch, and the two that fail most often.
How AI answer engines retrieve and cite sources, and what actually makes a roofing site citable.
Find out if your site is citable.
$2,500, and you see every answer a competitor is cited for and you are not.