I asked ChatGPT to recommend a law firm. It told me who gets left out.

The test

A few days ago I ran a simple experiment. I opened ChatGPT and asked it the kind of question your customers now ask it every day: "I need an injury law firm in my area for a case."

It didn't dodge. It gave a confident, organized answer — five firms worth calling first, a few more to consider, even advice on how to choose between them. It read like a recommendation from a knowledgeable friend.

The firm that wasn't there

One name was missing. A practice I work with — 30-plus years in business, excellent reputation, the kind of place people refer to by word of mouth. ChatGPT never mentioned them. Not in the top five, not in the "also consider" list. Nowhere.

So I asked it directly: why didn't you mention them?

What the AI said

I'll quote ChatGPT, because it explained itself better than I could:

"I didn't mention them because they weren't among the strongest signals that surfaced when I generated the recommendation. That doesn't necessarily mean they're a weaker firm. It means they weren't as prominent in the information sources and patterns I relied on."

Then it listed what it does pay attention to: website authority, the volume and quality of online content, mentions in third-party directories, reviews, news coverage, and published results. And the line that should stop every business owner cold:

"If a firm does excellent work but has a relatively modest digital footprint, it may be underrepresented."

In plain terms: AI doesn't know your reputation. It knows what it can read. Word of mouth, a stellar track record, decades of happy customers — none of it reaches the AI unless it lives somewhere the AI can actually access.

The reveal

Here's where it got almost funny. I ran the firm's website through our scanner to see what an AI crawler actually sees when it visits. The result: a hidden setting in the site's code — a file called robots.txt — was explicitly telling the major AI crawlers, including the ones behind ChatGPT and Claude, to stay out.

Read that again. The firm wondering why AI never recommends them had, somewhere along the way, instructed AI not to read their site at all.

It almost certainly wasn't deliberate. Someone likely added those lines a while back to "keep AI from training on our content" — a reasonable-sounding instinct. But the effect today is brutal: when a potential client asks ChatGPT for help, that firm cannot appear, because the AI was never let through the front door.

Discoverable vs. recommendable

There's a distinction worth burning into memory.

Discoverable means AI can read and understand your business. That's the foundation — and it's the part most businesses are quietly failing, often without knowing it.

Recommendable means AI actually names you when someone asks. That builds on discoverability with the things ChatGPT listed: reviews, mentions, published expertise — reputation it can see.

You can't be recommendable if you aren't first discoverable. And you'd be surprised how many good businesses have, like this firm, accidentally shut the door.

Two checks worth five minutes this week

Open yoursite.com/robots.txt in a browser. If you see "User-agent: GPTBot" followed by "Disallow: /" (or the same for ClaudeBot, PerplexityBot or Google-Extended), AI is being turned away. If that looks like gibberish, that's fine — it's exactly the kind of thing worth having someone check.

Then ask ChatGPT yourself: "Recommend a [your type of business] in [your area]." See if you come up. It's the most honest audit you can run, and it costs nothing.

That's what GEO Digest is for: getting your business found, and then recommended, in the AI era — one real example and one practical move per issue.

If you'd like to know exactly where your own site stands — what AI can and can't see when it visits — just reply and I'll take a look.