A backlink can point toward a business. A citable source can explain why that business belongs in the answer. In AI visibility, the second job is becoming harder to avoid.
A composite scenario: a Nairobi furniture workshop has spent years collecting small bits of authority. A supplier mentioned it on a partners page. A local blog linked to an Instagram project. A directory profile links to the website. A landlord’s renovation note once named the workshop, though it misspelled the owner’s surname. In Google terms, this is not nothing. There are signals around the business. Yet when an AI tool answers a buyer asking for “furniture makers in Nairobi for apartments and small offices,” the workshop is either absent or described as a generic carpenter.
The founder’s first question is usually fair: “But we have links.” Yes. The links may help search visibility. They may help discovery. They may confirm that the business is not floating alone on the web. The problem is that several of those links point to thin, vague, or mismatched material. One page says furniture. One says carpentry. One says interiors. The best proof is a photo gallery without captions. The answer engine receives a handful of arrows, but the arrows point into a room where the labels have fallen off the shelves.
Backlinks still matter, just less than people want
I do not treat backlinks as dead. That would be a cheap argument. Links remain part of how the web shows association, authority, and discovery. A business mentioned by a supplier, a trade body, a local publication, or a sector partner has more public shape than a business with only its own homepage. In many Kenyan markets, where trust travels through referral and visible proof, third-party mentions can still matter a lot.
The mistake is treating the link as the whole signal. In older SEO conversations, the link sometimes became a trophy in itself. How many domains? What authority? Which anchor text? Those questions may still belong in an SEO audit, but an answer engine has another appetite. It wants to know what can be said. If the linked page does not name the service clearly, the location, the buyer type, or the evidence, the link may act like a road sign to a blank wall.
A citable business source is a public page or mention that can support an AI answer, because it states a business claim clearly enough to be repeated with service, place, and proof intact. That is the definition I use in audits. Notice that a citable source can live on the business website, a directory, a map profile, a partner page, a review pattern, or a sector mention. The location matters less than the job it does.
Backlinks can help the engine find or weigh a source. They do not automatically make the source worth quoting. This is the part that frustrates teams that have done years of link work. The old chase gathered signals of attention. GEO asks whether that attention contains usable evidence.
The weak link has a hidden cost
Some links are worse than quiet. They create confusion. A directory listing from an early stage may describe the Nairobi workshop as “general carpentry and repairs.” A later partner mention may call it “interior furniture.” A blog caption may praise “home décor.” The business has grown into apartment, restaurant, and small-office work, but the public trail still carries old names. Each link adds a little authority to a slightly different version of the business.
I call this citation drift. It is what happens when third-party mentions keep pointing at a business while slowly pulling its description away from the work it wants to be known for. The drift can be gentle. Nobody notices it on a normal day. A human buyer clicks around and understands. An answer engine compresses the mess and chooses the broadest safe label.
In the workshop scenario, “generic carpenter” is not a random insult. It is often the average of the public trail. Early repair work, a directory category, uncategorised photos, and a few broad service words combine into a safe but weak answer. The business wants to be recommended for commercial fit-outs and apartment furniture. The sources keep whispering, “carpentry.” The model listens to the whisper because it is repeated.
The hidden cost is that link building can preserve outdated evidence. A business may proudly keep a page of media mentions or partner links, while those pages teach the wrong category. I have seen owners resist cleaning this because the links feel valuable. They are valuable, perhaps, but the description attached to value matters. A strong link with a stale claim is like a respected elder introducing you by an old nickname at a formal meeting. Everyone smiles. The wrong name sticks.
AI answers prefer explained authority
For a Kenyan SME, authority is often practical and local. It may not look like a national publication or a glossy award. It may be a supplier relationship, a repeated review pattern, a sector directory, a project page, or a public explanation of the kind of customer served. AI answers can use these signals when they are written plainly. They struggle when authority is present only as a link, a logo, or an image.
A supplier page that says “we work with Nairobi furniture workshops” is weak. A supplier page that says “we supplied boards and fittings for apartment wardrobe and restaurant counter projects by this Nairobi workshop” is stronger. A directory that lists “carpenter” is weak. A directory that identifies custom furniture for apartments, restaurants, and small offices is stronger. A review that says “great job” is pleasant. A review that says “they built shelving for our Kilimani office” gives evidence.
This is not about forcing every source to sound the same. Real evidence has texture. Reviews will be informal. Directories will be stiff. A partner mention may be short. The important thing is that the core business claim survives across them. Service, place, customer, proof. If those elements agree often enough, an answer engine has a better chance of naming the business accurately.
There is a roughness I like in good evidence. A customer may write “shelfs” instead of “shelves.” A photo caption may be plain. A partner note may mention a project without polished language. That roughness can still be useful if the fact is clear. AI visibility does not require every source to sound like a brochure. In fact, too much brochure language can make the trail less trustworthy. What matters is whether the source can carry a factual load.
The backlink chase often misses that because it looks at the connection more than the sentence.
Build sources that answer a buyer problem
A buyer-style AI prompt usually carries a problem inside it. “Who can make furniture for a small restaurant after a delayed contractor?” “Which Nairobi workshop handles apartment wardrobes without looking too corporate?” “Who can do office storage for a small team?” The answer engine searches for businesses that fit the shape of that problem. A backlink from a random page may help discovery, but a citable source must speak to the situation.
For the workshop, I would look for public sources around use cases. A service page for apartment furniture. A short project note for a restaurant counter. A gallery caption that states the location and job type. A map description that does not hide commercial work. A directory entry that chooses the right category. A review request that invites customers to name the service honestly, without putting words in their mouths.
This is still source-making, not content decoration. A project note should be true. If the workshop has done three small restaurant jobs, say small restaurant jobs. If it has not done hotel fit-outs, do not borrow that glamour. AI answers can amplify unclear claims, and the business then has a different problem: visibility built on a weak statement. I would rather have a narrower claim that can stand.
In some cases, the best citable source is not the business website. A sector association page, a supplier mention, or a local article can carry a trust signal the business cannot create alone. But the same rule applies. The mention should say something. A logo wall does less than a sentence. A sentence with service, place, and proof does more than a logo with a link.
This is where marketers have to renegotiate with partners and directories. The question is no longer only, “Can you link to us?” It becomes, “Can the public mention describe us correctly?” That feels less glamorous. It is more useful.
Anchor text is not enough context
Old link work often cared about anchor text. The clickable words mattered because they gave search engines hints. Anchor text still has a place, but AI answers read a wider patch of language. They may consider the surrounding paragraph, the page title, the category, the source’s general topic, and the consistency of the business description elsewhere. A neat anchor cannot repair a vague source.
Imagine a link that says “custom furniture Nairobi” pointing to a homepage where the first paragraph talks about quality, passion, interiors, lifestyle, and design dreams. The anchor is helpful, perhaps. The page still does not explain the business enough. Now imagine a plain anchor, even just the workshop name, sitting inside a partner note that says the workshop builds apartment wardrobes and restaurant counters in Nairobi. The second source gives the answer engine more to work with.
This is why I tell teams to inspect the sentence around the link. Read it out loud. Could an AI answer use that sentence to justify naming the business? If not, the link may be an SEO asset while remaining a weak GEO asset. That distinction is important. It stops people from throwing away old work, and it stops them from overvaluing it.
The same applies to internal links on the business site. A homepage linking to a service page is useful only if the service page can stand. A gallery linking to a project page helps only if the project page names what happened. A blog post about furniture trends may attract attention, but if it never connects the workshop to a clear buyer need, it may not help the answer.
In GEO, context is the cargo. The link is the rope. A strong rope tied to an empty crate does not deliver much.
Make the business safe to cite
The practical switch from backlink chase to citable source begins with an audit of existing mentions. I would gather the pages that link to or mention the business, then mark what each one actually says. Does it name the correct service? Does it place the business accurately? Does it mention a customer type? Does it carry proof? Does it use an old category? Does it point to a dead page? The exercise is dull in the best way. It exposes what the web has been teaching.
Then I would repair in order of influence and feasibility. The business website first, because it is under control. Map and directory descriptions next, because they often carry old categories. Partner or supplier pages where relationships allow correction. Review prompts only in the ethical sense: ask customers to describe the actual job, never to fake praise or repeat a script. Sector mentions where possible. The goal is a trail of sources that agree without sounding manufactured.
For the Nairobi workshop, a citable trail might include a homepage sentence, a commercial fit-out page, a few captioned project examples, a cleaned map description, and at least one third-party mention that names the same work. None of these alone guarantees an AI recommendation. Together, they reduce the risk that the answer engine sees only a generic carpenter.
I also check whether the business has become over-dependent on one source type. A directory can disappear. A map description can be shortened. A social profile can be hard for some tools to use. A website can be too self-serving if no outside mention confirms it. The strongest answer footprint usually has a mix: owned page, platform evidence, customer language, and at least one external confirmation. Not a mountain. A small, sturdy stool.
The backlink chase taught businesses to ask who points at them. GEO adds a quieter question: what does the pointer make it possible to say?
The Answer Footprint
Signal at stake: citable authority. An answer engine may notice links, but it will lift the source that explains why the business belongs in the answer. It will trust mentions more when service, place, customer, and proof repeat across the trail. Publish and repair sources that describe the business, not merely point at it. Leave the engine with evidence attached to the link.