AI Strategy 12 min read

SEO and GEO Checklist for SA Business Websites (2026)

By ·

More than half of the web requests on Cloudflare’s network now come from bots, not people: 57.4% automated against 42.6% human by mid-2026 (NBC News, 4 June 2026, retrieved 6 October 2026). A growing share of those bots are AI assistants reading websites on behalf of someone who wants to hire a contractor, find a supplier or compare engineering firms.

That makes two jobs for a business website. It still has to rank on Google (SEO). It now also has to be read, understood and quoted by ChatGPT, Gemini, Perplexity and Google’s own AI answers (GEO, generative engine optimisation). The good news is that the two overlap far more than the hype suggests. This SEO and GEO checklist walks through ten steps, grouped into four questions, that we apply to every website we build and that I applied to our own site in September. If you only have a week, the one-page version near the end tells you where to start.

Key Takeaways

  • SEO and GEO share the same foundation: pages machines can read, clear answers, consistent business details and visible trust signals.
  • The one GEO-specific technical decision is your crawler policy. Each AI company runs separate crawlers for search and for training, and blocking the wrong one can remove you from AI answers.
  • Google says no special markup or AI text file is needed to appear in AI Overviews or AI Mode; a page only has to be indexed and eligible for a snippet.
  • You can measure AI visibility for free with a fixed list of prompts asked monthly and a referrer filter in your analytics.

What SEO and GEO Have in Common

SEO makes your pages rank in a search engine’s list of results so that people click through. GEO makes your business easy for an AI assistant to find, understand and cite when it writes an answer. Both depend on the same four things: a site that crawlers can read without running scripts, content that answers specific questions clearly, business information that is consistent everywhere it appears, and reasons to trust you such as named people, real projects and reviews. The differences are in emphasis and measurement: GEO cares more about self-contained passages that can be quoted accurately, and it is measured in citations and enquiries rather than rankings and clicks. For a fuller comparison, see our SEO and AI search-ready websites page.

The most thorough checklist we found while researching this post, a 4,500-word guide from the SEO platform Incremys (incremys.com, updated April 2026), organises the work into technical, content, authority and monitoring tracks. That structure is sound and we follow a similar one below. Where it falls short for a South African business is in the details: it never separates AI search crawlers from training crawlers, it offers no robots.txt lines you can copy, and it has nothing to say about Google Business Profile, WhatsApp, prepaid mobile data or POPIA. Those gaps are what this version fills.

Before You Begin

You will need:

  • Access to your website’s files or CMS, or your web developer on call. Steps 2, 6 and 9 need someone who can change the site’s code; the rest you can do yourself.
  • A Google account for Google Search Console and Google Business Profile, and a Microsoft account for Bing Webmaster Tools.
  • Your business details written down once: legal name, trading name, phone number, email, physical or service area, and registrations.
  • About two hours for the checks, and a few days for the fixes.
  • Difficulty: the checks are easy; three of the fixes (Steps 2, 6 and 9) need a developer.

Track 1: Can Machines Read Your Site?

Step 1: Decide your AI crawler policy in robots.txt

By the end of this step you will know exactly which AI systems can read your site and which cannot. Most businesses have never decided; their robots.txt is whatever the website builder or a security plugin wrote.

The important fact is that each AI company runs more than one crawler, for different purposes, and they are controlled separately. According to the companies’ own documentation:

CompanySearch and answer crawlers (allow if you want enquiries)Training crawlers (your choice)User-triggered fetchers
OpenAIOAI-SearchBot surfaces sites in ChatGPT’s search featuresGPTBot crawls content that may be used to train modelsChatGPT-User acts when a user asks; robots.txt may not apply
AnthropicClaude-SearchBot improves search results for usersClaudeBot collects content that may contribute to trainingClaude-User fetches pages when a user asks
PerplexityPerplexityBot surfaces and links sites in Perplexity results; not used for foundation-model trainingNone documentedPerplexity-User generally ignores robots.txt because a user requested the page
GoogleGooglebot indexes Search, which AI Overviews and AI Mode draw onGoogle-Extended controls use for Gemini training and grounding; Google states it does not affect Search inclusion or ranking—

Sources: OpenAI, Anthropic, Perplexity and Google, all retrieved 6 October 2026.

Three kinds of AI crawler and how to treat eachA diagram with three columns: search crawlers such as OAI-SearchBot, Claude-SearchBot, PerplexityBot and Googlebot, labelled allow if you want enquiries; training crawlers such as GPTBot, ClaudeBot and Google-Extended, labelled your choice; and user-triggered fetchers such as ChatGPT-User, Claude-User and Perplexity-User, labelled may ignore robots.txt.Three kinds of AI crawler, three different decisionsSearch crawlersOAI-SearchBotClaude-SearchBotPerplexityBotGooglebotAllow if you wantenquiriesTraining crawlersGPTBotClaudeBotGoogle-ExtendedYour choice; blockingdoes not hide youUser fetchersChatGPT-UserClaude-UserPerplexity-UserAct for a person;may ignore robots.txtBased on OpenAI, Anthropic, Perplexity and Google crawler documentation, October 2026

To check your site, open yourdomain.co.za/robots.txt in a browser. Look for Disallow: / under any of the search crawlers above, or under User-agent: *, or for a security plugin’s “block bad bots” list that sweeps them up. For a business that wants to be found, a sensible policy allows the search crawlers explicitly and makes a separate, conscious decision about training. A minimal version looks like this:

User-agent: *
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

# Training crawlers: allow or disallow as a deliberate choice
User-agent: GPTBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: Google-Extended
Allow: /

Sitemap: https://www.yourdomain.co.za/sitemap-index.xml

Two notes. Blocking Google-Extended does not affect Google Search, AI Overviews or AI Mode, by Google’s own statement, so it is purely a training decision. And the user-triggered fetchers cannot reliably be blocked with robots.txt, because they act on a person’s request; the practical answer is to make sure what they fetch is accurate.

Verification: load the file in a browser and confirm the search crawlers are allowed. Then test the site in Google Search Console’s URL inspection tool to confirm Googlebot can fetch a key page.

Step 2: Make sure your content is actually in the HTML

This step tells you whether crawlers see your pages the way people do. Google renders JavaScript before indexing; most AI crawlers are not documented as doing so, and the safe assumption is that they read only what arrives in the first response. It catches subtle cases too: when I audited our own site in September 2026, the main heading and introduction on seven pages were set to fade in with a scroll animation, so a crawler that ran no scripts saw an empty hero section.

  1. Open a key service page in Chrome, press F12 for developer tools, open the command menu (Ctrl+Shift+P or Cmd+Shift+P), type “Disable JavaScript”, select it, and reload the page.
  2. Whatever disappears is what a non-rendering crawler never sees. On many site-builder and heavily scripted sites, that is the whole page.
  3. Alternatively, right-click and choose “View page source”, then search for a sentence from your main heading. If it is not there, it is being injected by script.

The fix is server-side rendering or a static build, where every page is delivered as complete HTML. That is a rebuild decision for most site-builder sites, which is why it sits in the “if your site fails most of this” section below.

Verification: your main heading, introduction and contact details are visible with JavaScript disabled, and present in the page source.

Step 3: Fix the basics every crawler relies on

By the end of this step, search engines know your site exists, which pages matter, and when they changed.

  1. HTTPS everywhere, with the plain http:// and non-www addresses redirecting permanently to one canonical version. When we audited our own site in September 2026, three address variants returned the page separately instead of redirecting; it is an easy thing to miss.
  2. An XML sitemap with real last-modified dates, referenced from robots.txt and submitted in Google Search Console and Bing Webmaster Tools, Microsoft’s equivalent, which most businesses forget.
  3. No broken links and no soft 404s. Search Console’s Pages report lists them.
  4. Reasonable speed on a phone. Core Web Vitals in Search Console will tell you if real visitors are waiting.

Verification: Search Console shows your key pages as indexed, and the sitemap report shows them discovered.

Track 2: Is There Something Worth Quoting?

Step 4: Rewrite your main service pages answer-first

After this step, your most important page gives an AI assistant a self-contained passage it can quote accurately. This matters more than any technical setting: an assistant that can read your site but finds only “quality solutions tailored to your needs” has nothing to work with.

For each core service, the first paragraph after the heading should state, in plain words, what the service is, who it is for, where you offer it, to what standard, and roughly how the process works. Then use the rest of the page for detail. Compare:

  • Before: “We are a leading electrical contractor committed to excellence.”
  • After: “We are a registered electrical contractor based in Randburg, serving homes and small businesses across northern Johannesburg. We do fault finding, distribution board upgrades, backup power and solar installations, and we issue Certificates of Compliance for property sales. Most call-outs are same-day.”

The second version answers “who does electrical work near me”, “who can issue a CoC” and “do they do solar” in one passage. Give every service its own page rather than one long list; a page per service can rank, and be cited, on its own.

Verification: read the first paragraph of each service page aloud and ask whether a stranger could repeat what you do, for whom and where. If not, rewrite it.

Step 5: Add the specifics only you know

By the end of this step, your site contains facts that no competitor and no AI model already has. Assistants cite sources that add something: a named project, a capacity, a lead time, a credential, a service area. Generic text is the part of the internet that is already full.

Depending on your trade, that means:

  • Projects, with sector, location, scope and outcome, in two or three sentences each.
  • Credentials with dates: ISO certification, B-BBEE level, CIDB grade, NHBRC, PIRB or Department of Employment and Labour registration, PV GreenCard, professional registrations. We cover which ones matter on our pages for engineering firms and for contractors and trades.
  • Capacities and turnaround: sizes you handle, typical lead times, emergency availability.
  • Areas: the suburbs, towns or provinces you actually serve, on a small number of genuinely different area pages rather than hundreds of copy-paste ones.

Verification: count the specific, checkable facts on your homepage and main service page. Fewer than ten is a sign the site is talking about itself rather than informing a buyer.

Step 6: Structured data that matches the visible page

Done properly, this step lets machines read your business details without guessing. Structured data (JSON-LD) describes your organisation, services, locations and content in a format search engines and AI systems can parse.

Keep expectations honest. Google states that no special structured data, AI text files or markup are required to appear in AI Overviews or AI Mode; a page only needs to be indexed and eligible to show a snippet (Google Search Central, updated 10 December 2025). Structured data still earns its place because it removes ambiguity: the right business type (Electrician, GeneralContractor, ProfessionalService and so on), one consistent name, phone number and address, service areas, and links to your other profiles.

  1. Use the most specific business type that fits.
  2. Make every value match what is visible on the page. Mismatches undermine trust rather than build it.
  3. Mark up services, articles and authors; add FAQ markup only where the questions are genuinely useful to readers, because it no longer earns anything in Google’s results: Google limited FAQ rich results to government and health sites in 2023 and stopped showing them altogether on 7 May 2026 (Google Search Central).
  4. Validate with Google’s Rich Results Test or the Schema Markup Validator.

Verification: the validator shows no errors, and the name, phone and address in the markup are identical to the ones on the page and on your Google Business Profile.

Track 3: Can They Trust You?

Step 7: Be the same business everywhere

The aim of this step is simple: every place a machine can learn about your business says the same thing. AI assistants draw on several sources at once, so a website that calls you one thing, a Google listing that says another and a directory with an old phone number is a reason to cite someone else.

  1. Google Business Profile: claimed, verified, correct category, service areas, hours, and the same name, phone and website as the site.
  2. Bing Places: the same details, so Microsoft’s search and assistant products have them too.
  3. South African directories and trade bodies: the industry bodies and directories that matter in your trade, each with identical details. These third-party mentions are also the most likely way for AI systems to learn that your business is real and established.
  4. Social profiles linked from your structured data, so the connection is explicit.

Verification: search your business name and phone number; every result should agree.

Step 8: Show the people, the dates and the proof

By the end of this step, a sceptical reader, human or machine, can see who stands behind the site.

  1. Named people: the owner or lead on the about page, with a photo and relevant registrations; articles with a named author and a date.
  2. Reviews on platforms you do not control, chiefly Google. Ask at the moment a job is finished and send the link by WhatsApp. Testimonials typed into your own site carry little weight on their own.
  3. A last-updated date on pages that change, and an annual refresh of anything with a year in the title.
  4. A POPIA-compliant contact form and privacy page. The Protection of Personal Information Act requires you to say what you collect and why, to collect only what you need, and to keep it secure. A form that asks only for a name, number and message, states its purpose, and links to a plain privacy page covers the basic requirements of the Act and doubles as a trust signal.

Verification: your about page names a real person, your Google listing has recent reviews, and your contact form explains what happens to the data.

Track 4: South African Realities and Measurement

Step 9: Build for phones on prepaid data, and for WhatsApp

This step makes the site work for the way South Africans actually contact businesses. In our experience most visitors to South African trade and engineering sites arrive on a phone, often on prepaid data and sometimes in an area with patchy coverage, and many would rather message than fill in a form.

  1. Keep pages light: compressed images, no autoplaying video, no oversized scripts.
  2. Put a WhatsApp click-to-chat link and a tap-to-call number on every page, above the fold on mobile.
  3. Make quote forms short, and have them open WhatsApp with the details filled in rather than sending an email that waits until morning.
  4. Check the site on a cheap Android phone, not only on a desktop.

This is also where GEO pays off: an assistant that recommends you will usually hand the person your phone number or website, and the next step has to be effortless.

Verification: on a mid-range Android phone over mobile data, the homepage loads in a few seconds and WhatsApp opens from one tap.

Step 10: Measure it without expensive tools

This last step gives you a baseline for AI visibility and a way to see whether it changes. The paid AI-visibility tools are built for large brands; a small business can get most of the value for free.

  1. A prompt panel. Write down ten questions your buyers would ask, in their words: “who refurbishes industrial valves in Secunda”, “plumber in Centurion who replaces geysers on weekends”. On the first of every month, ask each one in ChatGPT, Gemini, Perplexity and Google, and record whether your business appears, what is said about it and which source is cited. Keep the spreadsheet.
  2. Referrals in your analytics. In Google Analytics, filter sessions by referrer for chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. The numbers will be small; the trend is what matters.
  3. Search Console and Bing Webmaster Tools for impressions and clicks, including the questions you are appearing for.
  4. Enquiries. Count WhatsApp messages, calls and quote requests, and ask new enquiries how they found you. This is the number the rest of the checklist exists to move.

A note on why clicks undersell the effect: in July 2025 Cloudflare measured Anthropic’s crawlers reading about 38,000 pages for every visitor they referred, OpenAI’s about 1,091, Perplexity’s about 195 and Google’s about 5 (Cloudflare, 29 August 2025). The assistant often answers without sending anyone through, so the enquiry, not the visit, is the signal.

Pages crawled per visitor referred, by AI platform, July 2025Horizontal bars on a logarithmic scale: Anthropic about 38,000 pages per referral, OpenAI about 1,091, Perplexity about 195, Google about 5.Pages read for every visitor sent back (July 2025)Anthropic38,066OpenAI1,091Perplexity195Google5.4Logarithmic scale. Source: Cloudflare blog, 29 August 2025. Ratios change month to month.

The One-Page Version: Ten Things to Do This Week

#Do thisTimeWho
1Open your robots.txt and confirm OAI-SearchBot, Claude-SearchBot, PerplexityBot and Googlebot are not blocked10 minYou
2Reload a service page with JavaScript disabled; note what vanishes15 minYou
3Submit your sitemap in Google Search Console and Bing Webmaster Tools30 minYou (developer if there is no sitemap)
4Rewrite the first paragraph of your top service page answer-first1 hourYou
5Add three projects with sector, place and outcome1 hourYou
6Validate your structured data and fix mismatches1 hourDeveloper
7Make Google Business Profile, Bing Places and your site say the same thing1 hourYou
8Add a named person and date to your about page; request two Google reviews30 minYou
9Put WhatsApp click-to-chat and tap-to-call on every page30 minDeveloper
10Write your ten-prompt panel and run it for the first time45 minYou

If a developer maintains your site, three questions get the technical half done: “Are our AI search crawlers allowed in robots.txt?”, “Is our main content in the HTML without JavaScript?” and “Does our structured data match the page?”

If Your Site Fails Most of This

Be honest about the result. If Steps 1, 2 and 6 all fail, and the site runs on an old page builder or a heavily customised WordPress install, patching it is often slower and dearer than rebuilding it properly. A modern static or server-rendered site delivers complete HTML to every crawler, carries its structured data by design, loads quickly on prepaid data, and is cheap to host. Every website we build now ships this way, with a custom CMS so you can edit it yourself and with your existing content moved across and redirected, so nothing you have earned on Google is lost.

If you are not sure which side of that line you are on, the free AI Visibility Scorecard takes a few minutes and tells you where your site stands on exactly these points. If you want the reasoning behind the whole approach, start with why bots now outnumber people online.

Frequently Asked Questions

What is the difference between an SEO checklist and a GEO checklist?

Most of it is the same list. Both need pages that crawlers can read, clear structure, consistent business details and trustworthy content. GEO adds three things: a deliberate decision about which AI crawlers to allow, content written so that a self-contained passage answers a question accurately, and measuring citations and enquiries rather than only rankings and clicks.

Should I let AI bots crawl my website?

If you want enquiries, allow the search crawlers that put you into AI answers, such as OAI-SearchBot, Claude-SearchBot, PerplexityBot and Googlebot. Whether to allow the training crawlers, such as GPTBot, ClaudeBot and Google-Extended, is a separate choice. Each company documents these as separately controlled crawlers, and Google states explicitly that blocking Google-Extended does not affect Google Search.

Do I need an llms.txt file?

No. Google states it does not use AI text files or special markup for AI Overviews or AI Mode, and the other assistants have not committed to reading llms.txt. It costs little to add and does no harm, but fix crawler access, page content and business details first.

Indirectly. Google says no special structured data is needed to appear in AI Overviews or AI Mode. Structured data still helps search engines and AI systems understand who you are, what you offer and where you work, and it removes ambiguity about your business name and contact details, so keep it accurate and matched to the visible page.

How long does it take for GEO to work?

Technical fixes are picked up as crawlers revisit your pages, usually within days to a few weeks. Being cited for competitive questions depends on how much specific, trustworthy information exists about your business across the web, so expect gradual change over months and check monthly with the same set of prompts.

Conclusion

Setting a website up for SEO and GEO is mostly the same work done with more care: let the right crawlers in, put your content in the HTML, say specifically what you do and where, be the same business everywhere, show the people behind it, and measure enquiries. None of it needs an expensive tool, and most of the checks take an afternoon. The businesses that will be recommended by AI assistants over the next few years are the ones that made these decisions deliberately while their competitors left them to a plugin.

References and Retrieval Notes

All sources retrieved 6 October 2026.

Interested in what we’re building?

Get in touch to learn more about our projects or discuss how we can help your business.