A growing share of people now ask ChatGPT, Claude, Gemini or Perplexity the questions they used to type into Google. Those systems recommend businesses. Yours is either part of those answers or it is not, and unlike traditional search there is no results page to appear lower down on.
This is what actually influences that, minus the hype.
First, check whether you have blocked them
Start here, because nothing else matters if the answer is wrong.
Many websites block AI crawlers without their owners knowing. Some content delivery networks and security platforms now offer one-click AI bot blocking, and some enable it by default. Hosting providers occasionally add it during setup.
Visit yoursite.com/robots.txt and look for entries such as GPTBot, ClaudeBot, Google-Extended, PerplexityBot, CCBot, Applebot-Extended or Bytespider. If those are disallowed, no amount of optimisation will get you cited, because the systems cannot read your site at all.
This is a genuine business decision rather than a purely technical one. Blocking protects your content from being used for model training. Allowing makes you discoverable in AI answers. You can also take a middle position, since some crawlers are used for live retrieval and citation rather than training, and those are the ones that affect whether you get recommended.
Whatever you choose, choose it deliberately. We regularly find businesses investing in content while silently blocked from the systems they hoped would surface it.
Why AI systems quote some pages and not others
When one of these systems answers a question, it retrieves material and summarises it. Some pages are far easier to use that way than others.
Specific beats general
“We deliver exceptional results for our clients” is unusable. It contains no fact worth repeating. “Most business websites take three to five weeks from approved brief to launch” is quotable, because it answers something.
Marketing language is the single biggest obstacle. Superlatives, mission statements and phrases like “cutting-edge” carry no information, so they get discarded. Numbers, timeframes, prices, constraints and clear definitions survive.
Answers positioned where they can be found
Put the answer immediately after the question, in the first sentence or two, then elaborate. Content that builds slowly to a conclusion buried in paragraph nine is far less likely to be extracted correctly.
Structure that signals what a section contains
Descriptive headings, short paragraphs and lists help. A heading reading “How long does a website take to build” is more useful than one reading “Timelines”, because it matches how the question is asked.
Content that admits limits
This is counterintuitive and it matters. Pages that explain when something is not a good fit read as more trustworthy and get cited more readily than pages claiming universal suitability. Saying “a progressive web app is the wrong choice if you need deep hardware access” is more citable than claiming PWAs suit everyone.
What about structured data?
Worth being accurate here, because there is a lot of overconfident advice.
Google has stated there is no special structured data requirement for its AI features, though any structured data you use should match your visible page content. Microsoft has said schema helps its systems understand content. Independent studies have found mixed evidence on whether schema coverage correlates with citation rates.
The reasonable position: implement Organization, Service, Article and Breadcrumb markup because it is well understood, cheap, and helps traditional search regardless. Do not expect it to be the lever that gets you cited. Content quality matters considerably more.
One point of genuine importance: your structured data must describe content that actually appears on the page. Markup describing invisible content is a policy problem, not a shortcut.
Being mentioned elsewhere matters more than people expect
These systems do not only read your website. They draw on the wider web. Directory listings, industry publications, reviews, forum discussions and news coverage all contribute to whether a business is recognised as a real entity.
Practically, that means consistency pays. Your business name, description and contact details should match everywhere they appear. A business described three different ways across the web is harder to recognise as one entity than a business described consistently.
The llms.txt question
You may have read that you need an llms.txt file, a plain-text summary of your site aimed at language models.
Honest assessment: it is a proposed convention, not a standard, and adoption by major AI systems is not established. It costs very little to publish and may help. Treat it as a low-cost experiment rather than a requirement, and be sceptical of anyone selling it as essential.
What actually seems to work
- Confirm you are not blocking AI crawlers, or that blocking is a decision you made deliberately.
- Answer real questions in plain language, using the words customers use rather than industry vocabulary.
- Publish specifics. Prices, ranges, timeframes, processes, constraints. These are what gets repeated.
- Say when your service is not the right fit. It builds trust with readers and reads as balanced to a summarising system.
- Keep your entity details consistent across your site and every listing that mentions you.
- Add sensible structured data that matches your visible content, without expecting miracles.
- Keep content current. Dated, stale pages are less likely to be treated as reliable.
What to be sceptical about
This field attracts confident claims built on very little evidence. Nobody outside these companies knows exactly how citation decisions are made, and the systems change frequently. Anyone offering guaranteed AI visibility, a proprietary ranking factor list, or a fixed number of citations is overselling.
The durable advice is unglamorous and looks a lot like good content practice: be specific, be honest, be structured, be findable.
What to take away
Check your robots.txt first, because blocked sites cannot be cited at all. Then write content containing facts worth repeating, structured so answers are easy to locate, including honest statements about limitations. Add reasonable schema, keep your details consistent across the web, and treat anything promising guaranteed results with suspicion.
If you would like this assessed on your own site, it is part of how we handle AI integration and website development. You can get in touch and we will start with the robots.txt check, which is free and takes a minute.
