An AI Discoverability Checklist for Your Website

Jake Hodges · August 28, 2026

More buyers now ask AI assistants for recommendations before they ever open a search engine. Who handles this kind of work. Which vendor fits. What to ask before signing. If ChatGPT or Claude cannot read and understand your website, your business is not part of those answers.

Here is the part most articles skip. Nobody outside the AI labs knows exactly how these systems choose their sources. Anyone promising guaranteed placement in AI answers is guessing.

But there is a short list of things that plausibly help, cost almost nothing, and improve your site for human visitors anyway. This work is often called GEO, short for generative engine optimization, which just means getting found by AI assistants. Here is the checklist we run.

How AI assistants read your site

Two paths matter. First, AI models learn from web pages collected by crawlers over time. Second, most assistants now search the web and read pages live before answering a question.

Both paths hit the same gate. Can the system reach your site, and can it pull clear facts out once it does. Every item below widens that gate.

The checklist

1. Let AI crawlers in

Your robots.txt file is a plain text file at yoursite.com/robots.txt that tells automated visitors what they may read. Many sites block AI crawlers without realizing it. Some security and hosting tools block them by default, and some owners flipped a switch years ago and forgot.

Look for rules affecting the major AI user agents. GPTBot from OpenAI. ClaudeBot from Anthropic. PerplexityBot. Google-Extended, which governs whether Google may use your content in its AI products.

Blocking is a fair choice for a publisher who sells content. For a service business that wants to be found, it works against you. Decide on purpose, not by accident.

2. State what you do in plain sentences

An assistant reading your homepage should be able to answer three questions. What does this company do. Who is it for. Where does it operate.

A surprising number of websites fail this test. The homepage leads with something like "Elevating brands through transformative experiences" and never names the actual service.

The fix costs nothing. Put one plain statement near the top of your homepage and every service page. We are a what, in where, doing what, for whom.

Then test it. Paste the page text into an assistant and ask what the company does and who it serves. If the answer comes back vague or wrong, rewrite until it does not.

3. Keep your business facts consistent everywhere

AI systems cross-reference. Your website, your Google Business Profile, your LinkedIn page, and the directories that list you all describe the same entity, meaning the cluster of facts that identifies your business. Name, services, locations, contact details.

When those sources disagree, you create doubt about basic facts. Doubt is a reason to cite someone else.

Make a list of everywhere your business appears online. Use the same official name, the same service descriptions, and the same locations in all of them. Boring work. Worth doing once and maintaining.

4. Add structured data

Structured data is machine-readable labeling embedded in your pages, usually as JSON-LD, a small block of code that states facts in a standard format. The vocabulary comes from schema.org.

At minimum, mark up your organization or local business, your core services, and any FAQ content. Search engines have read this format for years. A system parsing your page gets unambiguous facts instead of guessing from prose.

This usually takes a developer a few hours, and free validators confirm it is wired correctly. If your site was built without it, that is a small fix, not a rebuild.

5. Answer real questions, plainly

Assistants answer questions. Pages built as direct answers give them something to work with.

Collect the questions prospects actually ask on sales calls and in emails. Write each one as a heading. Answer it in the first one or two sentences. Add detail after the answer, not before it.

The pattern to avoid is the long warmup, where the real answer hides at the bottom of the page. Humans hate that too.

6. Add an llms.txt file

llms.txt is an emerging convention. It is a plain Markdown file at yoursite.com/llms.txt that summarizes what your site is about and links to your most important pages, written for AI systems rather than people.

We will be straight about this one. It is not an official standard, and nobody has proven the major assistants read it. It also takes less than an hour, cannot hurt, and states your key facts in the clearest possible format.

Think of it as a business card left at the front desk. Cheap, polite, possibly useful.

What not to expect

None of this guarantees your business shows up in AI answers. The systems change constantly, and the companies that build them publish little about how sources get chosen. Treat any vendor promising guaranteed AI visibility the way you would treat a guaranteed first-place Google ranking.

The claim we will stand behind is smaller. Every item on this list removes a reason for an AI system to skip you or misread you. Every item also helps traditional search and human readers. The cost is a few hours of focused work.

Run the list once, then recheck it quarterly. Crawler names change. Tools update their defaults. Your services evolve.

Where this fits

Discoverability sits on top of a website that works. If the site is slow, confusing, or untracked, being found is only half the problem. That is why we treat this as one layer of our Websites & Search work, alongside conversion and measurement.

If you want a second set of eyes, our AI discoverability audit checks every item above plus a few that require tooling. It pairs well with a broader website audit when the whole foundation is in question.

And if you are not sure which layer is the real bottleneck, the Performance Audit is our front door. We look at how your marketing actually runs, website included, and hand you a dollar-weighted plan you keep whether or not you hire us. Start with a Diagnostic Call.

The operational letter

One useful idea, when we have one.

No schedule, no filler. When we write something worth your inbox, you get it. Unsubscribe any time.

Want this handled instead of read about?