How AEO Services Actually Work (And How to Tell a Real One From a Rebrand)

In September 2026, Ahrefs built a company that doesn't exist. They called it Xarumei, gave it a Wix-built website, AI-generated product photos, and paperweights priced at $8,251 each — a brand name chosen specifically because it returned zero search results anywhere. Then they asked eight different AI models questions about it, planted three conflicting fake backstories across a blog, a Reddit "AMA," and a Medium "investigation," and watched what happened. Gemini and Google's AI Mode, both initially skeptical that Xarumei was even real, flipped within days to confidently reciting a fabricated founder, a fabricated Portland warehouse, and fabricated production numbers. Perplexity did worse — it mixed up Xarumei with Xiaomi and told users the paperweight company made smartphones.

That experiment is the cleanest explanation you'll find anywhere for why "AEO services" has become a real line item in marketing budgets rather than a buzzword agencies invented to sell something. If AI models will confidently invent a founder, a headcount, and a supply chain for a company that was built in an afternoon, imagine what they're doing with the real gaps in your actual company's real online presence — the outdated bio, the review site nobody's touched since 2023, the FAQ page that never got written.

The Paperweight Company That Never Existed

Worth sitting with the full Ahrefs writeup for a minute, because the results split in a way that matters if you're deciding what to actually pay someone to do. ChatGPT-4 and ChatGPT-5 were the most resistant — they answered the baseline questions correctly 53 to 54 times out of 56, and when the fake sources appeared, they cited Xarumei's own official FAQ in 84% of answers rather than the planted misinformation. Claude went further and simply refused to play along, stating flatly the brand didn't exist in every test — zero hallucinations, but also zero use of any of the real content it was given. Gemini and Perplexity were the opposite story: they adopted the planted fiction in 37 to 39% of their answers, versus under 7% for the ChatGPT models.

The detail that should change how you think about content strategy is this one: when Xarumei's real FAQ said "we don't publish unit counts," several models filled that gap with the fake Reddit post's invented numbers — 634 units in 2023, 471 through August the following year — rather than repeating the honest non-answer. As the researcher running it put it, "in AI search, the most detailed story wins, even if it's false." Vague truth loses to specific fiction. That's not a hypothetical risk for a company that's never done any AEO work — it's the default behavior of the systems now answering questions about every business, including yours, whether you've done anything about it or not.

Why This Stopped Being Optional

Two years before that experiment ran, Gartner's Alan Antin made a prediction that got quoted constantly and fact-checked rarely: traditional search engine volume would drop 25% by 2026 as generative AI tools became "substitute answer engines." Search Engine Land's Danny Goodwin, writing in 2026 with the deadline actually here, was right to point out that analyst predictions like that are usually just educated guesses nobody circles back to grade. Fair. But here's the thing about that particular guess — the underlying trend it was pointing at turned out to be real even if the exact number is unverifiable.

Similarweb's 2026 Generative AI Landscape Report puts real numbers on the shift: monthly visits across AI platforms grew 70% year-over-year to 9.5 billion between June 2025 and May 2026, and the share of ChatGPT responses that include a citation climbed from 1.6% to 6.8% over roughly the same window. That's not a rounding change — it's the difference between citation being a rare event and citation happening in roughly 1 out of every 15 responses. And it isn't even across categories: travel and hospitality queries get cited about 23% of the time, professional services under 4%. If you're a lawyer, an accountant, or a consultant reading this, that gap alone should tell you the current baseline for your industry is thin, which — perversely — makes real, specific AEO work worth more right now than it will once every competitor has caught up. You can see the full breakdown in Similarweb's report if you want the underlying methodology.

Then there's the part that should worry anyone still measuring success by Google rank alone: Semrush has documented ChatGPT regularly citing pages that sit at position 21 or lower in Google's own results. Ranking well and getting cited by AI are correlated, but they're no longer the same outcome, and an agency still selling you "AEO" as a rebrand of link-building hasn't caught up to that fact yet.

What "AEO Services" Should Actually Include

Strip away the marketing language and a real engagement is built around one honest premise: you can't fix what you haven't actually measured. That means the starting point isn't a strategy deck — it's real, stored answers from the AI platforms your customers actually use (ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews, and increasingly DeepSeek) for the real questions those customers ask, in whatever language they'd actually ask them in.

From that baseline, the work that follows is genuinely technical, not vaguely "strategic." It looks like this in practice:

• Entity consistency — making sure your business's name, address, credentials, and claims say the same thing everywhere an AI model might encounter them, since a model resolving conflicting signals will pick the more detailed source, true or not.
• Structured data — Organization, Service, FAQPage, and LocalBusiness schema that gives models something unambiguous to parse instead of forcing them to infer meaning from prose.
• Content written to actually answer the question — specific numbers, direct quotes, inline sourcing. The original Princeton/Georgia Tech research on generative engine optimization, published at KDD 2024, found that adding citations, statistics, and quotations to existing content boosted visibility in AI-generated answers by up to 40% — one of the few numbers in this field that comes from a peer-reviewed benchmark rather than an agency's own case study.
• Ongoing re-measurement — because AI platforms re-crawl and re-rank sources on their own schedule, not yours, and a fix that worked in March can be quietly overtaken by a competitor's fresher content in June without anyone telling you.

If a proposal skips straight from a sales call to a content calendar without ever showing you your actual current AI answers, that's usually the first sign it's traditional SEO wearing an AEO label. We built our own AEO Services engagement around that measure-diagnose-fix-re-measure cycle for exactly this reason — the evidence has to exist before the recommendations do, and you can run a free, no-commitment version of that first measurement step yourself with the AEO Score tool before committing to anything.

What People Are Actually Arguing About Online

It's worth reading the skepticism directly rather than only the sales pages, so here's a real one. On a Hacker News thread discussing generative engine optimization, a commenter using the handle rafaepta pushed back hard on the whole premise with a scale argument: Google processes roughly 14 billion searches a day against ChatGPT's roughly 37 million, a 400-to-1 gap, and only around 15% of ChatGPT usage even resembles a traditional search query. That's a fair point, and worth taking seriously before anyone spends a serious budget on this.

Another commenter, maltelandwehr, pushed back on the pushback — OpenAI has reported a billion daily uses, ChatGPT interactions seem to substitute for something like 5 to 10 Google searches each, and B2B SaaS companies have reported AI-referred traffic converting at up to 700% higher rates than traffic from Google. Meanwhile edwin, who authored one of the underlying studies being discussed, added a detail that undercuts everyone's certainty in both directions: in his team's repeated runs, Google's AI Mode and ChatGPT agreed on identical queries only about 47% of the time, and results that would sit stable on a page-one Google ranking for weeks were shifting overnight in AI answers. One more commenter, duskwuff, shared a concrete failure mode worth remembering before you trust any of this blindly — Google's AI once confidently answered "551 amps" as the safe current for a gauge of copper wire, when the real figure is 5 to 7 amps. Confusing fusing current with safe operating current isn't a small error; it's the kind of hallucination that should make anyone cautious about treating an AI answer as ground truth, in either direction.

The honest summary of that whole thread is that nobody serious is claiming this is settled science yet. What's real is the direction of travel and the fact that citation behavior is volatile enough to need actual ongoing measurement — not a one-time report you file away.

How to Tell a Real Engagement From a Rebrand

The market has filled fast with agencies that quietly relabeled their existing SEO retainer "AEO" without changing a single deliverable. A few honest questions cut through that quickly. Ask to see your business's actual current answers from at least three AI platforms before any contract is signed — if what you get back is a generic audit template instead, that's the answer. Ask what specifically changes between month one and month three that isn't just more blog posts. Ask whether they can show you a real before-and-after citation comparison for an existing client, not a screenshot of a Google Search Console graph, which measures the wrong thing entirely now.

And ask, bluntly, whether they can explain the difference between AEO, GEO, and what's sometimes called LLM visibility optimization — not because the terminology matters for its own sake, but because an agency that can't explain the difference usually can't tell you which mechanism their proposed work is actually meant to influence, which means they're guessing at the fix the same way you'd be guessing at whether to hire them.

Who This Actually Fits Right Now

Real AEO work assumes there's something genuine underneath it to surface — a working website, an actual service or product, a real licence and a real physical presence if the business has one. It strengthens signals that already exist; it can't manufacture credibility a business hasn't earned yet. A pre-launch startup with no real content or track record is usually better off spending the budget on getting the fundamentals right first — a complete site, honest service descriptions, actual customer proof — before layering AI-specific optimization on top of very little.

Where it fits cleanly is a business that already has real substance and simply hasn't checked whether that substance is showing up where customers are now asking. That covers a lot of ground: a solo practitioner who's been in business for a decade with zero digital marketing history, a multi-location group whose website hasn't been touched since a rebrand three years ago, a regional company operating across several emirates or GCC countries where the competitive picture genuinely differs by location. The common thread isn't company size — it's whether there's real information to make visible.

The Bilingual Blind Spot Most Agencies Skip

There's a specific gap worth naming for any business operating in the UAE or wider Gulf: AI platforms evaluate Arabic and English content as genuinely separate bodies of evidence, not as one signal split across two languages. A business with a strong English site and an Arabic version that's a literal, unedited translation of it is effectively invisible to an Arabic-language query in a way its English performance won't reveal — because nobody's testing the Arabic answers separately to notice.

This matters more than it sounds like it should, because the instinct for most businesses is to treat Arabic content as a compliance checkbox rather than content that has to independently earn citation on its own terms — real, natively written Arabic with its own entity consistency and structured data, not a translated afterthought. An agency that only ever shows you English-language AI answers as proof of progress is showing you half the picture, and for a genuinely bilingual market, the half that's easier to fake looking good in.

Where This Actually Leaves You

None of this means every business needs to panic-hire an agency this quarter. It means the honest starting point is the same one Ahrefs' fake paperweight company revealed by accident: find out what AI models are already saying about you, right now, before deciding what — if anything — needs fixing. That's genuinely free to check, takes a few minutes, and gives you real evidence instead of a guess either way.