I spent two weeks reading the actual documentation. The real thing. Google's Search Central guides, OpenAI's crawler docs, Microsoft's materials on Copilot and AI search, academic citation pattern research, Gartner forecasts. All of it. What I found is that most of what's being said about AI search visibility is either vague, wrong, or a repackaging of basic SEO advice with new terminology stapled on. Which is frustrating, because underneath the noise there are a few things that are genuinely worth understanding.
Here's what I actually learned.
Before you can be found, recommended, or cited in any AI-generated answer, your website has to be reachable. This is not a metaphor. The systems that power AI search are crawlers - software that visits web pages, reads their content, and stores what it finds. If a crawler can't get to your page, you don't exist in that system's world.
Google's documentation spells out the baseline: pages shown in AI features must be indexed and eligible to appear in regular search results. No special AI eligibility. Just ordinary crawlability.
OpenAI adds one specific requirement that surprises people: their crawler, OAI-SearchBot, must not be blocked in your robots.txt file. If it is, your site will not appear in ChatGPT search answers. That's the whole rule. Block the bot, don't appear. Don't block it, you're eligible. Many sites are blocking it without meaning to, simply because their robots.txt was set up conservatively or hasn't been reviewed since the site was built.
This is worth checking before anything else. Not because it's complicated, but because everything else is pointless if this step is broken.
As of early 2026, Google AI Overviews appear in roughly 25% of all searches and 57% of long-tail queries. Long-tail means the specific, multi-word searches: "HR consultant for NGOs in Nairobi," "packaging design for small food business Kenya," "accountant who handles construction company tax returns." These are the searches your actual clients are making.
More than half of those searches now produce an AI-generated summary before any links appear. The summary cites sources. Sometimes it links to them. Sometimes it doesn't. Either way, the user's first answer is coming from Google's synthesis rather than from clicking through to your page.
The research finding that matters most here: most AI Overview citations come from content that's already ranking on page one organically. You cannot leapfrog into AI answers from position forty. Traditional search performance is the floor, not an alternative to AI visibility. The two are not separate strategies.
Google describes its AI Mode as using something called "query fan-out" : it issues multiple related sub-queries simultaneously to pull information from a wider range of sources. A page about brand strategy in Nairobi has a better chance of being cited if it also addresses adjacent questions: how long it takes, what it costs, who it's suitable for, what the process looks like. Narrow pages get narrow citations. Depth matters.
ChatGPT now processes over a billion queries a day. As of early 2026, it accounts for roughly 77% of all AI-driven website referral traffic. Its market share in the AI assistant category is around 81%. These are not predictions. They're current measurements.
About 31% of Gen Z users say they begin searches using AI platforms rather than search engines. If your clients skew younger- startup founders, young professionals, procurement officers at tech companies- this shift is already affecting how you're being discovered, even if you can't see it in your analytics yet.
OpenAI runs two separate crawlers: OAI-SearchBot, which is used to surface websites in ChatGPT's search feature, and ChatGPT-User, which is used when ChatGPT browses the web during a conversation. These serve different functions. Confusing them leads to misguided technical decisions. OAI-SearchBot is the one that determines whether your content appears in ChatGPT's search answers. That's the one to check for in your robots.txt.
Citation pattern research found something worth noting: 44.2% of all citations in AI-generated answers come from the first 30% of the source document. The opening of a page carries disproportionate weight. If your most important service page buries the actual description of what you do in paragraph four, you've already lost most of the citation potential before a reader even gets there.
Of all the platform documentation available, Microsoft's materials on AI search are the most straightforward about what makes content useful inside an AI-generated answer.
Their guidance emphasises: concise direct answers, clear headings, comparison tables, Q&A sections, and sentences that remain meaningful when lifted from context. They call these "precision signals" - content that helps an AI system understand exactly what a business does, who it serves, and under what conditions, with enough specificity to be confidently cited.
The practical test for a precision signal is simple. Take a sentence from your service page. If you removed everything around it and read it in isolation, would it still communicate something useful and specific? "We help businesses grow through strategic brand development" fails this test. "We design brand identities for East African manufacturing companies preparing to expand into export markets" passes it.
South African businesses that have made this shift in how they write their service pages — away from aspirational generalities toward specific, operational descriptions - tend to show up more often in AI-generated comparisons and recommendation lists. Not because they did something technical. Because they became easier to quote.
Pulling from the research and platform documentation together, the pattern that emerges is consistent:
Pages that get cited in AI answers tend to be crawlable without restrictions, indexed in traditional search, specific about what the business does and for whom, structured with clear headings and discrete answerable sections, supported by evidence that the claims are real, and written in language that retains meaning when a passage is extracted from its surrounding context.
None of this is exotic. It's disciplined execution of things that good web writing has always required. The difference is that the margin for vagueness has shrunk. A page that was passable before - ranking quietly somewhere on page two, getting occasional traffic from people who scrolled past the top results — may now be below the threshold that AI systems need to make a confident citation.
The businesses being cited in AI answers are not necessarily the biggest or the most well-known. In Ethiopia's growing digital economy, in Rwanda's tech sector, in Kenya's legal and professional services market, smaller businesses with well-structured, evidence-rich web content are appearing in AI-generated recommendations ahead of larger competitors whose websites are vaguer and thinner. The playing field is genuinely more level than it used to be. But you have to play.
"I need to write pages specifically for AI." You don't. Write for humans. The best pages for AI citation are the best pages for human comprehension. They're the same thing. Pages written for machine extraction at the expense of human readability will fail both audiences.
"Schema markup will get me into AI answers." Schema helps systems categorise what kind of page they're looking at. It doesn't make a vague page useful. A well-written, specific service page without schema markup will be cited more often than a generic page with extensive structured data. Do the writing first. Add schema after.
"SEO is dead, I need a new strategy." SEO is not dead. It's the foundation. You cannot gain AI visibility without first earning search visibility. The research is clear on this. Treat AI optimisation as a layer that sits on top of -not instead of - your existing SEO fundamentals.
The thread running through everything here is clarity. A business that explains itself well- in plain language, with real evidence, with a structure that helps readers find what they need - is the kind of business AI systems can work with. We've written about this in the context of brand clarity as a growth engine, and the principle is identical here.
If you're working through whether your site currently meets these standards, the website audit checklist we've written for Kenyan businesses covers the diagnostic process in detail- technical access, content quality, trust signals, local legibility, and conversion. Run through it before investing in new content or technical work.
How quickly will changes to my site affect AI search visibility?
There's no fixed timeline. If your pages are already indexed, improvements can begin to show effect within weeks as AI systems re-crawl and update. If there are crawl errors or indexation problems, those need fixing first. Think of it as a continuous process, not a one-time project.
Is Perplexity worth optimising for separately?
Perplexity accounts for roughly 15% of AI referral traffic. The same principles apply: crawlable, explicit, evidenced content. There's no unique Perplexity optimisation separate from what you'd do for Google and ChatGPT.
Does blocking Googlebot affect ChatGPT visibility?
Yes, indirectly. ChatGPT's search function relies partly on content that's indexable through standard web discovery. A page that's blocked from Google is likely inaccessible to ChatGPT's search too. These systems are more interconnected than they appear.
Should I create separate pages targeting AI search?
No. Create genuinely useful, well-structured service pages that serve human readers. AI systems will find those pages more useful precisely because they were written for comprehension rather than extraction. Separate AI pages are unnecessary and likely to produce worse results.
%402x.svg){kind=icon}
