Getting Found on ChatGPT: Limits and Checks

ChatGPT is not a search engine with a ranking you can climb, and the sooner that lands the sooner the work becomes tractable. There is no index position, no submission form, no verified-business program, and no setting that makes it mention your company.

What there is, when web browsing is active, is a retrieval step: it searches, fetches some pages, and writes an answer grounded in what it found. That step is the only part of this you can influence, and influencing it is not the same as controlling the output.

Two different machines, wearing the same interface

Answering from training. The model produces text from what it absorbed during training, which happened months ago and included whatever the web said about your business at that time — including things that were wrong then and are wronger now. You cannot edit this. It ages out on somebody else's schedule.

Answering with retrieval. The model runs a search, fetches results, and summarizes them with citations. This reflects the live web, which means it reflects work you did last week.

Almost everything written about "optimizing for ChatGPT" fails to distinguish these, which is why so much of it is unfalsifiable. Before you draw any conclusion from a ChatGPT answer about your business, check whether it cited sources. If it did not, you are reading a memory.

What ChatGPT can find about you

With retrieval active, it can generally reach:

• Your website, if it is crawlable and the answer is in text rather than inside an image or behind a script that never renders for a fetcher.
• Your Google Business Profile information, as it surfaces through search results.
• Review platforms, directories and aggregators.
• Local news, trade associations, licensing bodies and permit records where those are public and indexed.
• Anything else on the open web that names you, including things you would rather it did not.

What it cannot find

• Content only reachable behind a login.
• Anything communicated only in a PDF nobody links to, or only in an image.
• Anything that exists only on a platform that blocks crawlers.
• Facts nobody has published anywhere — including, very often, your actual service area, your actual hours, and half the services you offer.

That last category is where most local businesses lose. Not to a competitor with better tactics, but because the specific thing a customer asked about is not written down in plain language anywhere on the open web.

The work that actually helps

None of this is exotic. It is the same work that makes a business legible to any system, human or machine.

Say what you do, in the words customers use. If people call it "AC not cooling" and your site says "HVAC system diagnostics," you have created a translation problem for no reason. Write both.

Say where you work, specifically. "Greater Austin" is not a service area a retrieval system can reason about. A list of the towns and neighborhoods you genuinely cover is.

Put the answers in text on the page. Not in a hero image, not in a carousel that renders client-side and never resolves for a fetcher, not in a PDF. If a person browsing with images off cannot read it, assume a machine cannot either.

Make the site fetchable. Reachable without JavaScript execution, fast enough not to time out, not blocked in robots.txt by accident. I have found the accidental-block case more than once on sites I inherited, and it is the cheapest problem in this entire field to fix.

Keep your name, address and phone number identical everywhere. Not similar. Identical, character for character, including how you punctuate the phone number.

Add structured data that matches what is visible. Schema is a description, not a lever. Marking up services you do not offer or hours you do not keep is worse than having no markup at all.

Get real reviews, and answer them. Review platforms are heavily represented in the sources these answers cite. This is also the one item on the list that has independent value regardless of what any assistant ever does.

What does not help

• Paying for placement. There is no placement to buy.
• Stuffing "AI" into your page copy.
• Publishing high volumes of thin articles. This is now an active liability under Google's guidance on scaled content, and it does nothing for retrieval either.
• Any special schema type claimed to signal an assistant. It does not exist.
• Trying to get the model to "remember" you. It does not work that way, and the mechanism people imagine is not the mechanism that exists.

How to check whether any of it moved

Do not check once. Ask the same question three times over several days, signed out, in a private window, with the location set to the market you serve, and record whether sources were cited each time.

You are looking for the pattern, not the answer. Appearing once is evidence that inclusion is possible. Never appearing across ten runs is much stronger evidence, and it is the more useful finding, because absence is easier to establish than presence.

If it names you but the details are wrong, that is the best possible outcome of the test: it is concrete, traceable to a source, and fixable this week.

The honest summary

You cannot make ChatGPT recommend your business. Nobody can, and the people who say otherwise are describing a mechanism that does not exist.

What you can do is make sure that when it looks, there is something accurate to find — in text, on the open web, consistent across the sources it actually reads. That is a real and finite piece of work. It also happens to be the same work that helps customers who arrive by any other route, which is the main reason I think it is worth doing at all.