Google has stopped behaving like a neat list of blue links and started acting like a machine that wants to answer, summarise, and decide for you. That is bad news for anyone still building pages as if search were a spreadsheet game of keywords, backlinks, and a bit of luck. The old trick was to aim for the query. The new reality is harsher, because Google wants the entity, the context, the authority, and the human proof behind the words.
For South African businesses, this is not an abstract shift happening somewhere in Silicon Valley. It changes how a Cape Town agency writes service pages, how a Joburg law firm publishes guides, how a Durban e-commerce brand structures product content, and how a developer thinks about schema, entity signals, and trust. If your site reads like it was assembled by a content mill in a hurry, AI search will not reward the effort. It will extract what it can, ignore the rest, and move on.
Google is not just ranking pages anymore
The clean old model was simple enough. Google crawled the web, indexed pages, measured signals, and ranked the result that looked best for the query. That model still exists in the plumbing, but it is no longer the whole machine. Google now behaves more like an answer engine that stores, interprets, and displays information directly, often without handing the full click back to the publisher.
AI Overviews and Gemini-powered search make that obvious. The interface is no longer just, “Here are ten links, good luck.” It is, “Here is the answer I have assembled from the web, now stop asking me to do the librarian bit.” That changes the incentive structure completely. A page is no longer valuable only because it can rank. It is valuable because it can be quoted, interpreted, or used as source material inside a generated response.
The operator framing here is blunt, and it is not wrong to be blunt. Google is a listed company. It answers to shareholder pressure. It is not some open-source public utility with a moral duty to preserve the old web for sentimental reasons. Its search systems are being tuned for usefulness, retention, and commercial durability, not for the nostalgia of SEO consultants who miss the 2010s.
The pitch also ties the shift to Google’s deeper relationship with US state infrastructure, especially after Matt Cutts left for the United States Digital Service. From there, the story runs through support for US government departments, including defence and intelligence, via infrastructure as a service, secure hosting, and work with CISA to counter foreign threats and software vulnerabilities. On 20 January 2025, Donald Trump signed an executive order reorganising and renaming USDS as the United States DOGE Service, with Elon Musk put in charge and the U.S. DOGE Service Temporary Organization created alongside it. Whether you treat that as a political backdrop or a structural clue, the point is the same: Google is not operating inside a neutral vacuum. It is embedded in a world of power, money, compliance, and strategic priorities.
The keyword era is fading fast
Traditional SEO was built on a fairly mechanical idea. Pick the phrase, match the phrase, repeat the phrase, get links, rank page. That model has not vanished overnight, but it has become far less reliable. Google’s current systems are far better at detecting when a page is trying to game language instead of serving a real need.
Keyword stuffing is the obvious casualty. So is backlink chasing that looks manufactured rather than earned. Mass-produced AI pages built around template paragraphs and recycled search terms are also getting squeezed. If the page exists mainly to imitate relevance, Google has more tools now to see through it.
This is why so many sites are discovering that content volume is not the same thing as visibility. A thousand near-identical posts about “best accounting software in South Africa” do not make you authoritative. They make you look like a site that discovered automation and immediately forgot about taste.
The shift is from keywords to entities. Google is better at recognising brands, people, products, places, and relationships between them. It wants to understand who you are, what you do, where you operate, and why any of that should matter to a person asking a complex question. That is a very different game from sprinkling a target phrase across headings and hoping the algorithm gets sentimental.
E-E-A-T is doing more of the heavy lifting
Google’s ranking systems now lean harder on E-E-A-T, which is shorthand for Experience, Expertise, Authoritativeness, and Trustworthiness. The phrase gets thrown around so much it sometimes sounds like conference wallpaper, but the underlying idea is practical. Google wants proof that the page comes from someone who actually knows the subject.
That means original research matters. Unique case studies matter. First-person reviews matter. Real operational detail matters. A thin rewrite of five competitor articles does not. Neither does a fake sense of expertise built out of generic language and tidy claims.
For a South African SaaS company, this could mean publishing a breakdown of how a new onboarding flow affected trial-to-paid conversion in a local market, with actual numbers and assumptions. For a law firm, it could mean a proper guide written by a practitioner who has handled the issue, not a ghostwritten explainer padded with legal-sounding filler. For a retailer, it could mean honest product commentary from people who have used the thing, compared it, and seen where it breaks.
The key point is not that Google has become sentimental about authenticity. It has not. It has become dependent on content that is hard to fake. That is why first-hand experience carries more weight now. AI can imitate syntax. It struggles to imitate lived detail that actually costs something to produce.
Human content is still the fuel
There is a funny contradiction at the heart of all this. Google’s AI features depend on human work even as they reshape search in ways that can reduce direct clicks. The models and answer layers still need a live open web. They need fresh writing, current examples, real documents, product pages, technical explanations, and local context to index, synthesise, and cite.
If the web turns into a swamp of low-effort machine text, the machine gets worse. It has fewer reliable signals to draw from, fewer credible sources to cite, and less variation in how real people explain real problems. The result is the thing everyone already sees too much of, which is bland AI slop that sounds polished and says almost nothing.
People still want actual stories. They want the ugly part, the exception, the result that did not fit the neat plan. They want interpretation, not just summary. That is the space human content still owns, and it is why the companies that keep publishing real-world insight will keep feeding the very systems that now compete with them.
For businesses in South Africa, this is a useful opportunity if they are willing to do the work. Local information is often under-served, especially in technical niches. A Johannesburg fintech can publish better practical guidance than a generic US blog on compliance workflows, because the local context is different. A Cape Town web team can explain what happens when WordPress, caching, and third-party scripts fight in the real world, not in some abstract “best practices” slide deck.
What to do instead
If your current content workflow still looks like “find keyword, ask AI for draft, publish,” you are building for a search system that is already dying. A better workflow is slower, but it actually has a shot.
Build around entities, not just phrases
Start with the real things in your business.
Use clear references to:
- Brand names
- People and roles
- Products and services
- Locations and service areas
- Tools, systems, and integrations
If you run a business in Pretoria serving clients in Gauteng, say that plainly. If a page is about Shopify, WordPress, GA4, or a specific CRM, name it. Stop hiding behind vague language that could describe any company anywhere.
Schema markup helps here because it gives Google structure it can parse. It does not magically make weak content strong, but it makes strong content easier to understand.
A simple example:
“`html “`
That alone will not save a thin page. It will, however, help establish what the page is about and how the entity fits into the broader web of signals.
Publish things AI cannot fake well
There are still a few content types that machine text handles badly unless a human gives it a spine.
Prioritise:
- Case studies with actual numbers
- Before and after site audits
- Screenshots from real tools
- First-hand reviews of products and workflows
- Technical guides based on work you have actually done
- Comparisons that include trade-offs, not just features
A page that says a migration improved load time is weak. A page that says the site moved from 4.8 seconds to 1.9 seconds, after removing render-blocking scripts and fixing image delivery, is useful. That second version can be checked. It has texture. It has a reason to exist.
Write like someone who has touched the system
This is where most SEO content still fails. It sounds assembled, not observed.
Readers can tell when a paragraph has never been near a live site, a staging environment, a failed release, or a client who wanted “more traffic” but refused to fix the homepage. Use detail that proves contact with reality.
A useful guide should mention things like:
- Which plugin broke the page
- Which report was checked
- Which metric moved
- Which assumption turned out to be wrong
- Which part of the workflow was manual and annoying
That is what makes the page worth reading, and it is what makes it harder for AI to flatten.
The practical SEO job now
The job of SEO has not disappeared. It has changed shape.
The new goal is not to trick a ranking system into noticing you. It is to become the source that an answer system trusts enough to quote, summarise, or surface in a result. That means:
- Publishing fewer, better pages
- Writing for a known entity and a real audience
- Showing evidence, not just confidence
- Using schema and structured data sensibly
- Keeping the content alive with updates, examples, and corrections
- Putting actual human judgement into the workflow
For South African companies, there is another angle worth taking seriously. Local search behaviour is full of friction, mixed terminology, and platform gaps. That makes clear, grounded content more valuable, not less. If you explain a thing well, with local context and a real point of view, you are more likely to be the source that Google’s systems extract from when they need an answer that feels specific rather than generic.
The old SEO playbook was built for a web that rewarded volume and repetition. The current one rewards precision, evidence, and recognisable human experience. That is a better deal for readers, and a far harsher one for anyone still shipping AI text that sounds confident because it has no memory of being wrong.
