5 pages your product site needs before an AI will cite it

AI assistants don't recommend products. They quote pages.
If your site has nothing quotable, you're not in the answer, no matter how good the product is.

I've spent a lot of time lately asking assistants about small products, mine and other people's, and reading which URLs they cite. The pattern is boring in the best way: the sites that get quoted are the ones that already answer the question in a sentence a machine can lift without rewriting.

This post is the checklist I wish I'd had. It's not a growth hack. Google itself says there's no special optimization required for its AI features beyond normal, helpful, crawlable pages. That's the point: most product sites fail at the normal part.

Why "quotable" beats "optimized"

An assistant answering "what's a good tool for X?" has a few seconds and a pile of retrieved text. It's going to prefer text that:

• states what something is, plainly, near the top
• answers one question per paragraph
• doesn't need the surrounding page to make sense
• comes from a page that looks maintained by someone real

Marketing copy fails all four. "Reimagine the way your team works" can't be quoted as an answer to anything. So here are the five pages that fix that.

1. A homepage that says what it is in one sentence

The first visible sentence on your homepage should survive being pasted into a stranger's chat with zero context.

The shape is simple:

[Name] is a [category] that [does one thing] for [specific people].

A fictional example: "Ledgerly is an invoicing tool that turns timesheets into invoices for freelance designers." Not exciting. Extremely quotable.

Put the same sentence (or a close variant) in your <title>, your meta description, your social bios and your app store listing. Consistency is a signal; five different taglines are noise.

2. An FAQ that answers the questions people actually type

Not "Why is Ledgerly amazing?" but:

• Is Ledgerly free?
• Does Ledgerly work with Stripe?
• Can I use Ledgerly outside the US?
• How is Ledgerly different from a spreadsheet?

Each answer starts with the answer. "Yes, the free plan covers up to three clients" beats a paragraph of warm-up. If you want to go further, mark it up with FAQPage structured data. It won't guarantee anything, but it removes ambiguity about what's a question and what's an answer.

Where do the questions come from? Your support inbox, your sales calls, the search box on your docs, and the questions people ask about competitors.

3. A "how it works" page in plain steps

Assistants love numbered steps because users ask "how do I...?" constantly. A page that says:

1. Connect your calendar.
2. Pick the client.
3. Send the invoice.

...is far more citable than a looping hero video. Keep each step to one sentence, use the words your users use, and avoid screenshots as the only carrier of meaning. A model can't quote a PNG.

4. An about page with a real person behind it

When an assistant decides whether a page is trustworthy enough to cite, it has to lean on signals of who's behind it. A real name, a short bio, a way to contact you, and a date on things that change.

This isn't vanity. "Built by a two-person team in a garage" is a fine about page, as long as it's true and specific. An about page that's just a mission statement gives the model nothing to anchor trust to.

5. A machine-friendly index: sitemap, llms.txt and a changelog

Three small files that do a lot of quiet work:

• sitemap.xml: still the most reliable way to tell crawlers what exists.
• llms.txt: a proposed convention for a plain-markdown summary of your site aimed at language models. It's a proposal, not a standard, and not every crawler reads it. It costs you twenty minutes and forces you to write the one-sentence description from step 1.
• A public changelog: dated entries are evidence the product is alive. "Is X still maintained?" is a real question people ask assistants.

What I'd skip

• Hidden keyword blocks or text aimed only at bots. That's how you end up classified as spam, and it's the same trick prompt-injection attackers use.
• Ten near-identical landing pages per keyword. One good page beats ten thin ones.
• Gating your docs behind a login. If a crawler can't read it, an assistant can't quote it.

The afternoon checklist

Run this against your own site today:

• [ ] Can a stranger understand your homepage's first sentence pasted alone?
• [ ] Is that sentence identical across title, meta, bios and store listings?
• [ ] Does your FAQ have at least 8 questions in your users' words?
• [ ] Does every FAQ answer start with the answer?
• [ ] Is there a numbered "how it works" page with no image-only steps?
• [ ] Does the about page name a real person and a way to reach them?
• [ ] Do you have a sitemap, an llms.txt and a dated changelog?
• [ ] Is anything important behind a login or rendered only by JavaScript?

Takeaways

1. Assistants quote sentences, not brands. Write sentences that stand alone.
2. One clear definition, repeated everywhere, beats five clever taglines.
3. FAQs and numbered steps map directly onto how people ask questions.
4. Trust signals (a real person, dates, contact) decide whether you're cited at all.
5. None of this is a trick. It's the boring basics most product sites skip.

Your turn

Which page on your site would you be most embarrassed to see quoted word for word by an AI, and what would you rewrite first?

I write about practice, feedback and the AI-readable web while building NEXO, a daily learning game.