AI Search Only Feels New When the SEO Was Thin
Every so often, a new search format gets treated like a new sport. The uniforms change. And the scoreboard changes. People start arguing over initials. But the training plan? That’s usually a lot less mysterious than it looks.
Sprinting is a decent way to think about it. A 100-meter dash and a 200-meter dash aren’t identical events, sure. One has a different rhythm, a different feel and a different margin for error. Still, nobody trains for either one by ignoring starts, acceleration, leg strength and pacing. The event would feel “disruptive” in a hurry, if the athlete skipped those basics. AI search works the same way for many sites. If the foundation was solid, the shift feels like an adjustment. Every new interface feels like a crisis, if the foundation was shaky.
That’s the part a few sharp voices have been circling from different angles lately. Jono Alderson’s been blunt about how much of this debate gets smuggled into a branding contest. Mordy Oberstein’s pushed on the same pressure point from another direction, and Ross Hudgens has also made the case that the fundamentals still decide who shows up and who doesn’t. Different accents, same complaint: people keep acting as if AI search invented the need for clarity, trust and decent site structure. It didn’t.
A new label doesn’t fix an old habit of weak SEO.
That’s why the GEO versus SEO argument can get so noisy and so unhelpful. GEO, AEO, answer engine optimization, or whatever acronym the week delivers, all tend to pull attention toward terminology instead of the more annoying question underneath it: was the site ever built on solid SEO practice in the first place? If the answer’s yes, AI search often feels like a new distribution layer on top of familiar work. If the answer’s no, the team’s usually trying to patch over years of vague pages, tangled architecture, thin product copy and brand messaging that could describe anybody.
That’s also why some teams feel blindsided while others mostly shrug. The shrug comes from sites that already did the boring stuff well. Clean information architecture. Pages that say what they actually mean. Internal links that make sense. Consistent naming. Real expertise on the page instead of marketing fog machine prose. Those sites don’t suddenly become perfect in AI search, but they also don’t need a rescue mission just because the query box got smarter.
The disruptive feeling, then, is often a diagnosis rather than a discovery. AI search hasn’t made the old rules irrelevant. It’s made weak SEO easier to spot. When a site was built on thin content and fuzzy signals, the new environment exposes that fast. The transition looks less like a moon landing and more like a software update with a slightly annoying changelog, when the site had real substance.
That’s the practical reality check this article’s built around. In my view, if AI search feels like a foreign country, the problem may not be the border crossing. It may be that the site never learned the language of SEO in the first place.
The Training That Still Wins Is Mostly Old-School SEO
AI search has a fancy new interface, but it still rewards a lot of the same habits Google’s been pushing for years. Keep its site in decent shape and show a coherent brand story across its pages, both search engines and language models have a much easier job, if a business can explain itself plainly. That’s not glamorous. It won’t make for a thrilling conference slide. Save a lot of pain later, it does, however.
Google’s own AI optimization guide and helpful content guidance circle the same territory for a reason. They keep coming back to clarity, usefulness, and a site that doesn’t confuse the reader or the crawler. The exact wording changes, the acronym soup changes, and the audience changes, but the underlying demand stays stubbornly familiar: tell people what you do, make it easy to find, and don’t make the page feel like it was assembled from half a dozen unrelated ideas.
If a page needs a decoder ring, the machine will probably move on before the customer finishes squinting at it.
That simple rule applies to more than copy. Through the business, a healthy website gives search systems a clean path. Pages load without drama. Navigation makes sense. Service pages describe actual services. Contact details are easy to find. The About page sounds like the same company that appears on the homepage, the Google Business Profile, the LinkedIn page, and the rest of the places customers tend to check when they’re deciding whether to trust you with money. Both traditional search and AI systems have less room to get confused, when those pieces line up.
This is where a lot of SEO teams trip over their own shoelaces. They start with the keyword list and work backward, as if the business itself were a side note. It should run the other way around. The real company, the actual offer, the customers it serves, the geography it covers, the problems it solves, all of that should drive the SEO strategy. If a roofing company mainly handles storm damage in two states, given the site should say that plainly. The pages should reflect that instead of pretending it’s a Swiss Army knife because a spreadsheet suggested more search volume somewhere else, if a software firm sells one core product to finance teams.
That’s the part people often ignore because it sounds too ordinary. There’s no secret sauce in rewriting a homepage so it says what the business actually does. There’s no dopamine hit in cleaning up duplicate service pages, fixing messy internal links, or trimming vague copy that sounds impressive and means nothing. Yet those are the jobs that make entity optimization work at all. They also make generative engine optimization less of a buzz phrase and more of a practical workflow.
The same goes for brand consistency. Search engines have spent years trying to figure out whether a site’s about one thing or seven half-related things. Language models are doing a similar job, just with a different interface and a wider appetite for context. Both are looking for the most consistent, trustworthy explanation of who the business’s and what it offers. Don’t be shocked when the machine picks the wrong one, if your own site gives three slightly different answers.
So yes, AI search changes the surface a bit. But the core training still looks familiar: write clearly, keep the site healthy, make the brand easy to understand and let the business define the SEO plan instead of the other way around. Once that part’s in place, the next question becomes less about what your site says and more about what the wider web says back.
What Actually Changed in AI Search
The part that catches people off guard isn’t that AI search exists. It’s that the machine is no longer acting like a sleepy intern who only reads the homepage and maybe your About page if it’s feeling curious. AI systems pull from a wider trail: your site, your profiles, your mentions, your product pages, your local listings, your bios, your schema, and whatever else can be read cleanly enough to form a coherent picture.
That broader reading pattern changes the game for brands that got used to thinking about off-site work as a link-building contest with a side of brand mentions. For years, a lot of off-site SEO got flattened into one question: how many links did we get, and from where? Links still matter, of course, but AI search doesn’t stop there. It compares sources. And it notices mismatches. And it tries to assemble a brand entity from a stack of readable references rather than from a single shiny page.
AI search does not ask your homepage what you meant. It compares your site with everything else that says who you are.
That’s where entity optimization stops sounding like jargon and starts looking like basic housekeeping. If your business name appears one way on your site, another way in directories and a third way in third-party writeups, the model has to decide which version to trust. Same with services, categories, locations, founder names and product descriptions. A brand that says “paid search,” “PPC,” and “search ads” in different places without any clear pattern may still be understandable to a human. To a machine assembling a profile for Google AI Overviews or AI Mode, that sloppiness can make the picture blurrier than it needs to be.
That’s a real shift, but it’s not a mystical one. Search engines have cared about consistent entities for a long time. The difference now is that language models are much more willing to stitch together meaning from a larger set of documents. They don’t need every answer to come from your homepage. A review profile, a knowledge panel, a directory listing, a press mention, and your own structured data, they can combine signals from an FAQ. If those sources disagree, the model may not politely ask for clarification. It may just pick the version that seems most repeated or most credible.
And yes, this is where the old backlink-first mindset starts to feel a bit dated. Back then, off-site SEO often meant “get links from places that count.” That was never the full story, but it was the story many teams told themselves because it was easier to measure. AI search has less patience for that simplification. A backlink from a messy, inconsistent page does not magically clean up a brand entity. A mention from a trusted source can do more for understanding than a stack of irrelevant links ever did, especially when the machine is trying to decide whether two different names refer to the same company or two different ones.
This is also why structured data has not become optional just because AI can read more text. Clear markup helps machines parse what a page says, and Google’s structured data policies spell out the boundaries pretty plainly. Bing’s webmaster guidelines make a similar point in their own way: clean, honest signals beat clever tricks that confuse crawlers. None of that is flashy. It’s just the unglamorous part of making your brand legible.
For teams that already did solid SEO work, this probably feels less like a total reinvention and more like a stricter version of the same exam. Clear messaging. Consistent naming, and strong internal and external signals. A real business structure that can be understood by a machine without translation gymnastics. That’s familiar territory, just with a few more places for the machine to check your homework.
” It’s that it evaluates a wider set of proof points and expects those proof points to agree. That wider evaluation can feel brutal, if your site was built on thin content and loose brand definitions. If your foundation was already sound, the change’s there, but it’s not especially mysterious. The next question is the practical one: what does the model already believe about your brand, and where did it get that idea?
How to Audit What AI Believes About Your Brand
the audit becomes less mystical and more slightly tedious, once you accept that AI systems piece together a brand from lots of places. Which is good news, really, and you don’t need a crystal ball. You need a search window, a spreadsheet and a willingness to read your own copy with the suspicion usually reserved for a suspiciously cheap used car.
Start with the site itself. Check every major page, not just the homepage and the pretty sales pages the team likes to send to prospects. Read the about page, product pages, service pages, pricing pages, FAQ pages and any comparison pages you’ve published. Then check the supporting material that often gets ignored: author bios, contact details, location references, schema markup and footer language. If your site says one thing on the homepage, something softer on the about page and something completely different in a buried PDF, you’ve just handed a model three versions of the same company. It’ll pick one. You may not enjoy which one.
After that, move off-site. Look at your Google Business Profile, LinkedIn company page, YouTube descriptions, directory listings, partner pages, review profiles, trade association entries, and any third-party mentions that rank for your brand name. The question is not, “Do we have a presence here?” The real question is whether the facts stay stable from place to place. If your service list changes, your category changes, or your location data drifts, you create room for confusion. That’s where hallucination gets a foothold.
AI usually gets brand facts wrong for boring reasons: inconsistent wording, stale profiles, and too many versions of the same story.
This is where technical SEO stops being background noise and starts acting like a sanity check. If a page is blocked from crawling, noindexed, or tucked behind a setup that search systems can’t easily read, you need to know that before you blame the model for missing it. Google’s own documentation on robots meta tags and crawling controls is worth a look here, because visibility problems often begin with simple access problems. A model can’t learn much from a page it never properly sees.
You should also test the models directly. Ask plain questions about the brand, the products and the services. Try prompts like: What does this company do? Who’s it for? What are its main products? Where’s it based? What do people seem to misunderstand about it? Then ask a follow-up that forces the model to explain itself: Where did you get that information? Which pages or sources shaped your answer? Some models will cite sources cleanly. Others will mumble a little and move on. Either way, you learn something useful.
Don’t confuse that with a category ranking test. If you search a broad query like “best CRM for small teams” or “email marketing software,” you’re testing visibility in a market. That can matter, but it doesn’t tell you whether the model understands your brand. A brand-belief test asks something narrower and more revealing: what the model thinks your company is, what it sells, and whether it can keep those facts straight. A brand can show up in a category query and still be a mess when asked to describe itself. That’s a very different problem.
The best audit process’s less about hunting for one magical prompt where your brand appears and more about reverse engineering the path the model took to get there. Check whether that name still appears on a stale landing page, an old partner listing, or a directory entry nobody’s touched since 2023, if it repeats an old product name. If it gets your service area wrong, look for mismatched location data. Fair enough. If it describes your offer too vaguely, that often traces back to vague site copy. The model isn’t always inventing things out of thin air. Sometimes it’s just assembling a sloppy paper trail.
That’s why this kind of review’s so useful. It shows you where brand consistency’s strong, where it frays and where technical SEO work still matters in a very practical way. Fix the facts first. Then worry about whether the model’s paying attention for the right reasons.
The Bigger Payoff: Build for Accuracy, Then for Action
The same system that answers a question today is likely to do more than answer later. It may compare products, sort vendors, book appointments, fill carts, or trigger some kind of purchase path on a user’s behalf. That’s the part many teams keep half-noticing and then politely ignoring, like a smoke alarm with manners. The point isn’t just, “Can the model mention us?” It’s, “Can the model do something useful with the facts it has about us?”
If the machine gets your brand wrong on the way in, don’t expect it to make clean decisions on the way out.
That’s why the accuracy work comes first. Before anyone worries about agentic interactions, the model needs a clean picture of who you are, what you sell, where you operate, what you charge, and what makes your offer different from the next tab over. If that picture’s fuzzy, future actions become messy fast. A shopping assistant that can’t tell which products are current, which services are local, or which policy page’s authoritative won’t magically become competent just because the interface looks smarter than last year.
This is also where a lot of brands get tripped up. They treat AI search as a visibility problem when it’s often a data problem in a nicer outfit. If your site says one thing, your directory profiles say another and your product pages bury the useful details under marketing fluff, the model has to guess. Guessing’s fine for pub trivia. It’s a bad strategy for recommendations, and even worse for transactions.
Citations do matter. They help systems point to a source and reassure the user that the answer didn’t come out of a snack machine. Still, citations are only one part of the picture. A brand can earn a mention and still have a muddy identity. It can collect references and still confuse the system about location, category, pricing, or service scope. What the model needs is consistency it can read repeatedly across pages, profiles, feeds and public mentions. That consistency makes it easier for the machine to trust the brand, and for the user to trust the machine.
That is the bigger payoff here. Brands that clean up their representation now give themselves a better shot at showing up in future AI-driven buying flows without the awkward “wait, who are you again?” problem. The work is not glamorous. It’s closer to housecleaning than sorcery. But in AI search, that’s usually the difference between being considered and being skipped.
AI search only feels like a new sport to teams that were never training properly in the first place.
