How do I show up when people ask ChatGPT?

It searches, reads a few pages, and writes a paragraph

When a homeowner asks who to hire, whichever one they asked usually runs a search, opens a handful of pages, reads them and writes its answer from what it just read. The small links beside the answer point at those pages. Some of what it says can also come from what the model learned months earlier, which is how ChatGPT can know your name and still give out your old phone number.

Each company runs this its own way, and they do not share one index or one list of sources. OpenAI documents its own search crawler for ChatGPT, and in July 2026 Yelp and OpenAI announced a deal licensing Yelp's business data for ChatGPT. Google Search Central says its AI features rest on the same ordinary SEO fundamentals as the rest of Google Search (2026). A plumber can be well placed on Google and missing from the sources ChatGPT or Perplexity uses, or the other way round.

How Google and AI answers pick a business walks through the whole process.

Search robots and training robots

Each company sends named robots to read websites, and they do different jobs. The search robots, such as OAI-SearchBot, Claude-SearchBot, PerplexityBot and Googlebot, fetch pages so an answer can quote and link them. The training robots, such as GPTBot, ClaudeBot and Google-Extended, collect text for training models.

The crawler documentation from OpenAI, Anthropic, Perplexity and Google (2026) is clear that blocking a training robot does not take a site out of answers. Blocking a search robot shuts that engine out. A roofer who decided not to feed model training can keep that choice and still be readable for search. The mistake is blocking both without knowing it.

How blocked pages and script-only sites get skipped

Two things quietly shut ChatGPT, Claude and Perplexity out. The first is a block nobody chose: a line in robots.txt copied from a template, or a bot-protection setting at the host or security service that answers the robots with an error. The site loads fine for you and your customers, so nobody notices.

The second is a site whose words only appear after scripts run in a browser. Google renders those pages. A GEO audit dated September 6, 2026 found no evidence that the other AI fetchers run scripts, so a page built that way can reach them close to empty. That is one analysis, so we compare what a plain fetch returns with what a browser shows before calling any page unreadable.

A painter whose service area and phone number live inside a widget may be showing those facts to Google and nobody else.

What your own site has to say

ChatGPT writing a sentence about you can only use sentences it finds. Put the facts in text, near the top of the pages that matter.

• The trade and the jobs. An HVAC company should say heat pump installation, furnace repair, and ductless mini-splits, not "comfort solutions."
• The towns and counties. Written out: Vancouver, Camas, Battle Ground. Not only a map picture.
• The license and where to check it. An Oregon plumber holds both a CCB license and a BCD plumbing contractor license, and a plumber working in Clark County needs the Washington L&I plumbing contractor license.
• The proof. Years in business, warranty terms, photos of your own jobs, testimonials with a name and a neighborhood.
• Answers in your own numbers. What a water heater replacement usually runs, how long a tear-off takes, when a repair makes more sense than a replacement.

No special file or markup does this job for you. Schema, FAQ markup and an llms.txt file are not what makes ChatGPT recommend a contractor. Plain, specific words on pages the robots can read are the part you control.

Why outside sources differ by trade

ChatGPT, Claude, Gemini and Perplexity also read what other sites say about you, and which sites carry weight changes with the work. A roofer can be listed on the contractor locators run by GAF, Owens Corning, CertainTeed and Portland-based Malarkey. A painter has nothing like that: the big paint brands run no public find-a-painter tool, so Houzz and the Painting Contractors Association's Portland chapter do more of the work. A heating contractor can appear in the Energy Trust of Oregon Trade Ally network.

What matters most is that these sources agree with each other and with your site: same name, same phone, same license, same service area. With Yelp data licensed for ChatGPT since July 2026, an old phone number on Yelp is more than cosmetic. Bing Places is worth claiming for Bing and Copilot, but it is not a door into ChatGPT. Which directories matter goes through the list.

Ask twice and the names can change

Ask ChatGPT the same question twice and you can get a different list. Ask from another town, or type roof leak instead of roofer, and it changes again. One answer is not a ranking. A screenshot of your name in ChatGPT does not mean it will be there tomorrow, and a missing name does not mean it was missing yesterday.

A useful check asks many questions of each engine, counts each one separately, and reports how often a business was named, with the date. How long it takes covers what a fair comparison later looks like.

What the first report can show

A first read can show whether the search robots can reach your site, whether your pages say what you do and where, whether your listings agree, and whether you were named in the answers we ran on the day we ran them.

It cannot show why ChatGPT picked one company over another. Those companies do not publish that, and anyone claiming to know the formula is guessing. Nobody can honestly promise to put you in ChatGPT. What we can promise is the work that removes the reasons you are being left out, and a straight account of what the answers show afterward.