AEO for Kenyan Businesses: Getting Cited by ChatGPT and AI Overviews

Answer engine optimization (AEO) is the work of making your business legible to systems that answer questions directly: ChatGPT, Gemini, Perplexity, Copilot and Google's AI Overviews. It is not a separate discipline from SEO. It is what SEO becomes when the answer appears above the link, and the machine chooses which handful of sources to name.

For most Kenyan businesses, the practical consequence is uncomfortable. You can rank well and still be absent from the answer. You can also be cited without ranking, because an AI system assembles its response from sources that look authoritative for that specific question — not from a single ordered list of ten blue links.

This is a guide to the parts you can control, the parts you cannot, and the ones worth ignoring.

Why this is happening now, not next year

The numbers have moved quickly. Conductor's 2026 benchmark found that roughly a quarter of Google searches now trigger an AI Overview, up from about 13 percent a year earlier. EMARKETER expects close to a third of the US population to use generative AI search in 2026. In B2B, HubSpot's January 2026 survey of CRM buyers found 42 percent used AI search during their evaluation, and those buyers were more likely to purchase.

Kenya lags those figures, but not by much, and the direction is the same. The people evaluating software, suppliers and professional services in Nairobi are already asking ChatGPT to shortlist vendors. The answer they get is not a directory. It is a synthesis of whatever the model could retrieve and trust.

That has two effects. First, discovery increasingly happens inside a conversation where you are either mentioned or you are not. Second, the traffic that does arrive is further down the funnel, because the basic questions have already been answered. Both are reasons to care about how your business appears in machine-generated answers, and neither is solved by publishing more blog posts.

How an AI answer is actually built

It helps to stop thinking of AI search as a smarter search engine and start thinking of it as a research assistant with a deadline. The process has three stages, and each one asks something different of your content.

1. Retrieval

The system expands your question into several related queries, then gathers candidate sources: web pages, documentation, forums, reviews, news, video transcripts. If your content is not crawlable, not indexed, or not about the specific sub-question being asked, it never enters the pool.

What this rewards: clear crawl access, specific pages per question, and content that uses the vocabulary people actually use.

2. Selection

Candidates are ranked by relevance, authority, freshness and corroboration — whether other sources say the same thing. This is where a well-known brand with a vague page loses to a smaller site with a precise one. It is also where being mentioned elsewhere starts to matter more than the number of backlinks pointing at your homepage.

What this rewards: a clear entity, consistent facts across the web, and pages that look like they were written by someone who has done the work.

3. Synthesis

The model extracts passages, compresses them, and writes an answer with citations. Extraction favours short, self-contained statements: a definition, a number, a step, a comparison. Long introductions and clever framing get skipped because the model is looking for something it can lift without losing meaning.

What this rewards: answer-first structure, tables, explicit dates, and sentences that survive being quoted out of context.

The uncomfortable statistic

Analysis of AI citations has repeatedly found that only a small fraction of URLs cited by large language models also appear in Google's top ten results for the same query — one 2026 study put it at around 12 percent. Treat the exact number with caution; the direction is what matters.

It means two things. Ranking first is not a guarantee of being cited. And a page that ranks fourteenth can be the source an AI answer quotes, if it answers the question more directly than the page above it.

This is good news for smaller Kenyan firms. You are not trying to outspend a global brand on domain authority. You are trying to be the most extractable, most specific and most locally credible source for a narrow set of questions.

Seven levers you can actually pull

1. Answer the question in the first 60 words

Every page should open with a direct answer to the question in its title, before context, before the company story, before the clever anecdote. Forty to sixty words is enough. Then explain.

This feels blunt to write and is disproportionately effective. It also helps human readers who are scanning on a phone, which is most of them.

2. Be an unmistakable entity

AI systems need to know who you are, what you do, where you operate and whether you are the same organisation mentioned elsewhere. That means consistent naming, a physical address, a phone number that matches your Google Business Profile, a real About page with named people, and sameAs links to your LinkedIn, GitHub and other profiles.

Vagueness is expensive here. "A leading provider of innovative solutions" gives a retrieval system nothing to attach to. "A software engineering firm in Nairobi building custom applications, cloud infrastructure and production AI systems" gives it three entities and a location.

3. Make the structure machine-readable

Schema markup is not a ranking trick, but it removes ambiguity. At minimum: Organization with a URL, logo and address; Article or BlogPosting on editorial pages with author and dates; BreadcrumbList on inner pages; FAQPage only where a real FAQ exists; Service on service pages.

The rule is simple: mark up what is visibly on the page. Structured data that describes content the user cannot see is a liability, not an optimisation.

4. Let the crawlers in

You cannot be cited by a system that is not allowed to read you. Check robots.txt for accidental blocks, and decide deliberately about GPTBot, ClaudeBot, PerplexityBot and Google-Extended.

Blocking AI crawlers is a legitimate business decision if you sell content. If you sell services and want to be found, blocking them removes you from the answer entirely. A surprising number of Kenyan sites block by copying a template they did not read.

Also check the basics: server-rendered content rather than client-only rendering, clean canonical URLs, no accidental noindex on pages that matter, and a sitemap that is current.

5. Keep it fresh, visibly

AI systems weigh recency when choosing sources. Reported analysis in 2026 suggested pages updated within a month earned roughly three times more citations than older pages, and pages left untouched for a quarter were around three times more likely to lose citations they had.

In practice: show a published date and an updated date, refresh your most commercially important pages every quarter with new data or a new section, and do not pretend a cosmetic edit is an update. Changing the date without changing the substance is the kind of thing that erodes trust on both sides of the screen.

6. Earn corroboration off-site

Models look for agreement across sources. If your site says you are the best at something and nothing else on the internet says it, the claim is weak. If a industry publication, a review platform, a conference listing and a few practitioners say it, the claim becomes retrievable fact.

For Kenyan firms the highest-leverage corroboration is unglamorous: a complete Clutch or The Manifest profile with real reviews, a Google Business Profile with recent activity, LinkedIn posts from named employees, conference and community involvement, and being quoted in local tech press. None of this is link building in the 2015 sense. It is evidence building.

7. Be specific in ways a model cannot invent

This is the lever most businesses skip. Generic advice is everywhere, so it is cheap to retrieve and cheap to replace. Specific numbers, named tools, real failure modes, local regulations, actual costs and named sources are hard to synthesise and therefore more likely to be quoted.

For a Kenyan business, local specificity is a structural advantage. A page explaining how eTIMS reconciliation actually behaves, or what a Daraja callback timeout looks like in production, or how ODPC registration applies to a SaaS product, cannot be written by a global agency guessing. That is the content worth investing in.

What is different in Kenya

The fundamentals are universal. The evidence is local.

• Google Business Profile still matters. Local intent queries and AI local answers frequently draw on business profile data. Categories, hours, service areas and reviews are all inputs.
• Directories are citation sources, not just links. Clutch, The Manifest, Kenyan business directories and industry association lists are read by both buyers and retrieval systems. Inconsistent addresses and phone numbers create ambiguity.
• English dominates, but the gap is an opening. Most AI training and citation material is English. Well-written Swahili content on practical business topics is thin. If you serve a Swahili-speaking audience, that is an underexploited advantage — provided the quality is genuinely good, not machine-translated.
• Local media carries weight. Business Daily, TechCabal, CIO East Africa and specialist newsletters are the kind of sources a model can name. A mention there does more for citation likelihood than ten low-quality guest posts.
• WhatsApp shapes distribution. A large share of Kenyan business content travels through WhatsApp groups. If your article has no Open Graph image and a weak title, it arrives as a bare link and gets ignored.
• Regulation is a content moat. The Data Protection Act, ODPC guidance, KRA's eTIMS requirements and the Kenya AI Strategy are all under-covered relative to how often businesses need them. Precise, current explanations get cited.

What does not work

A short list, because most AEO advice currently being sold in this market is one of these.

• llms.txt as a silver bullet. It is a proposed convention, not an adopted standard, and major providers have not committed to it. Publish it if you like. Do not build a strategy on it.
• Publishing twenty AI-written articles a week. Volume without specificity gives a model nothing to prefer. It also creates near-duplicate pages that compete with each other.
• Keyword density. Retrieval is semantic. Repeating "software company in Kenya" eleven times does not help and reads badly.
• Fake FAQ schema. Marking up questions you do not answer on the page is a manual-action risk, not a shortcut.
• Gating everything behind a form. A PDF behind an email wall cannot be retrieved, quoted or cited.
• Blocking AI crawlers and then wondering why you are invisible. Pick one.
• Chasing a single answer. AI responses vary between sessions, models and even phrasings. Optimise for the pattern, not for one screenshot.

How to measure whether it is working

There is no equivalent of a rank tracker for AI answers, and anyone selling you one number is oversimplifying. What you can do is build a small, repeatable measurement routine.

1. A prompt set. Write 15 to 20 questions a real buyer would ask — "best software development company in Nairobi", "how much does custom software cost in Kenya", "how do I integrate M-Pesa into my app". Ask them monthly in ChatGPT, Gemini, Perplexity and Google. Record whether you are cited, mentioned, linked, and which competitors appear.
2. Referral traffic. Segment sessions from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com in analytics. The volumes are small and the intent is high.
3. Branded search. A rising volume of people searching your name is the clearest sign that off-site mentions are working.
4. Assisted conversions. Check whether AI-referred sessions appear earlier in journeys that later convert through direct or organic.
5. Source share. For a handful of target questions, note which domains get cited. That tells you which publications to pursue.

Expect a lag. Technical fixes can show up in weeks; citation share moves over months, and it moves fastest for pages that are refreshed and corroborated rather than merely published.

A reasonable first thirty days

1. Fix crawl and indexation basics: canonicals, robots.txt, sitemap, server-rendered content.
2. Rewrite the top of your five most commercially important pages to answer the title question in the first 60 words.
3. Add Organization, Article and BreadcrumbList schema with real, visible data.
4. Make your business details identical across your site, Google Business Profile and the directories that matter.
5. Pick three questions your buyers actually ask, and publish three specific, locally grounded answers — with dates and named sources.

That is unglamorous work. It is also the difference between having a website and being part of the answer.

Questions we get asked

Is AEO different from SEO?

The underlying work overlaps heavily: crawlability, structure, authority, freshness. AEO adds emphasis on extraction, entity clarity and off-site corroboration. Treat it as a change of emphasis, not a replacement.

Does llms.txt matter?

Not yet in any reliable way. It is worth understanding and cheap to publish, but it is not a substitute for crawlable, well-structured, specific content.

Should I block AI crawlers?

Only if your business model depends on people visiting your site to read content you sell. If you sell services, blocking removes you from the channel where your buyers are now researching.

How long does it take?

Technical fixes can affect retrieval within weeks. Meaningful citation share typically takes three to six months of consistent publishing and off-site corroboration. Anyone promising faster is guessing.

Can a small Kenyan business be cited?

Yes, and more easily than it can outrank a multinational. Specificity beats scale in retrieval. A precise answer about a local regulation, price range or integration is more useful to a model than a generic page from a larger brand.

If you want to know how your business currently appears across AI answers, we run a structured visibility audit: the prompts your buyers use, what each engine says, which sources it cites, and the specific pages to fix first.