Get the full AI search visibility guide

Chapter 01

How AI answers pick sources

Here’s the short version. Some answers come from what the model learned in training, with no links. When a question needs current information, ChatGPT searches the web, reads a handful of pages and writes the answer from them, with links to the ones it used.

Those cited pages are the part you can influence. Perplexity and Google’s AI Overviews work much the same way.

Ranking helps. It doesn’t decide.

Here’s a result I didn’t like. On 3 October 2026, one of our comparison posts was the first organic Google result for its question. ChatGPT answered the same question by citing the two platforms’ own pages, and didn’t name us.

Ahrefs saw the same pattern at scale. In a study of 75,000 brands (published 12 December 2025), branded web mentions tracked AI visibility far more closely than domain strength did:

Those are correlations with AI visibility (0 to 1), not proof of cause, and Ahrefs says so. But it matched what we saw on our own site: being talked about elsewhere mattered more than how strong our domain was.

Chapter 02

Let the right crawlers in

If the search crawler can’t reach your pages, nothing else in this guide matters. OpenAI’s crawler docs say sites that opt out of OAI-SearchBot won’t be shown in ChatGPT search answers.

The same split exists elsewhere: Anthropic’s Claude-SearchBot and PerplexityBot decide whether you can show up in those answers, while the training crawlers are a separate business decision. Blocking a training crawler doesn’t take you out of search answers.

Our own robots.txt lets every crawler read every public page and only blocks internal views:

User-agent: *
Allow: /
Disallow: /dev/
Disallow: /styleguide
Disallow: /industries/directory

Two things that catch teams out

• Bot protection on your CDN or firewall can block OAI-SearchBot even when robots.txt allows it. OpenAI publishes its IP ranges so you can allow them.
• Text that only appears after JavaScript runs. Many AI crawlers don’t run JavaScript, so text loaded by a script or hidden in a tab can be missed. Keep the answer in the HTML.

Chapter 03

One answer-first page per buyer question

Here’s the method. Write down the questions your buyers ask right before they shortlist: which tool, which one against which, what it costs. Map each question to one page on your site. If a question has no page, that’s your next page.

Then write each page so a sentence can be lifted as it stands:

1. Answer in the first two sentences. The question in the H1, the answer straight after it.
2. Use the buyer’s words in your headings. Phrase the H2s as the follow-up questions they ask next.
3. Put real specifics on the page. Prices, plan names, dates, customer names and numbers you can stand behind. “Affordable” gives an engine nothing to quote; “from $3,000 a month” does.
4. Sign and date it. A named author and an updated date tell readers, and engines, the page is maintained.

Comparison pages, pricing pages and “who is this for” pages matter most, because they match the questions people ask right before they pick.

Chapter 04

Get mentioned where the answers look

Here’s the part that’s easy to skip. On our first check, on 3 October 2026, every ChatGPT answer that named us cited our Clutch profile. None of our own pages was a source in those three answers.

So the answers lean on what other sites say about you. Where to work, in order:

• Review profiles. Clutch for services, G2 or Capterra for software. Keep services, prices and screenshots current, and ask happy customers for a review after each project or renewal.
• “Best tools” lists. Find the lists the answers cite for your category, then ask the authors to consider you, with the facts that make the case. Don’t pay for a link: Google treats that as link spam.
• Reddit and communities. Answer questions in your category as yourself, say who you work for, and skip the pitch.
• LinkedIn and YouTube. Founder posts and product videos give engines more places that describe what you do. YouTube had the strongest correlation in the Ahrefs study.

Before you start, ask your buyer questions and read the sources under each answer. If they already cite a review profile or a directory, fix that page first. It’s the cheapest win in this guide.

Chapter 05

Schema and llms.txt, honestly

Nobody has shown that structured data makes ChatGPT cite you. Google says structured data isn’t required for generative AI search, and there’s no special schema for it.

Here’s why we add it anyway: Organization JSON-LD with sameAs links ties your site to the profiles the answers already quote, and it’s a one-time job. Ours is on every page:

{
  "@type": "Organization",
  "name": "DesignMe",
  "sameAs": [
    "https://www.linkedin.com/company/designmeagency/",
    "https://clutch.co/profile/designme-design-no-code-experts",
    "https://www.youtube.com/@uiadrian"
  ]
}

Add Article schema with the author and dates on posts, and FAQPage schema where the page shows the questions and answers. Keep every claim in the schema the same as the page.

llms.txt

An llms.txt is a Markdown map of your site, proposed by Jeremy Howard on 3 September 2024. The honest part: nobody has shown it changes what ChatGPT cites, and Google says Google Search ignores the file.

We publish one anyway, because it’s cheap to keep right. Ours is generated from the same data as our pages, so it never drifts from what the site says. Treat yours as small upkeep, last on the list.

Chapter 06

Measure it every month

Results move, so the check is a routine, not a one-off. Here’s ours:

On our own site the scoreboard covers 55 buyer questions: 20 we search on Google every Monday, and 35 we put to AI assistants every month. Visits from AI assistants are counted every day.

ChatGPT tags the links it sends with utm_source=chatgpt.com, so those visits show up by source in your analytics. We counted 122 in the 28 days to 3 November 2026, almost all from ChatGPT.

Between checks, our free AEO checker scores a page on crawler access, structure and the signals AI answers rely on, with a fix for each gap.

Chapter 07

What we saw on our own site, and where to start

Here’s our baseline from 3 October 2026: the same ten buyer questions on three assistants. ChatGPT named us in 3 of 10 answers, Claude and Perplexity in 1 each. Every ChatGPT mention cited Clutch, none cited our site.

Then the next check, on 27 October: a bigger set (35 questions instead of ten) and partly different assistants, so read it as a new baseline, not a trend.

Across all three, 21 of 105 answers cite one of our pages and 7 name us. We can’t say which change did it, and the question sets differ. But our own pages now show up as sources, which they didn’t on 3 October.

And where we don’t show up yet: we’re in the Google top 10 for 50 of the 546 searches we track, and the top 3 for 22. Most of them aren’t on page one. That gap is our plan for the next pages we write.

Where to start

Run your homepage and your main pricing or comparison page through the AEO checker, fix what it flags, and write down your buyer questions this week.

If you want a team to do the fixes, the pages and the monthly check for you, that’s our SEO and AEO service, from $3,000 a month. And one honest line from us: nobody controls what ChatGPT says. We improve what it reads, measure what changes, and tell you when it doesn’t.