Key takeaways
- Guaranteed AI citations are not possible. Large language models are probabilistic, so identical prompts can return different cited brands from one run to the next.
- The churn is measurable. Research from Profound found 40 to 60% of cited domains change within a month for identical prompts, rising to 70 to 90% over six months.
- Around 80% of AI visibility is still fundamental SEO. Any GEO provider who doesn’t lead with that number is worth questioning.
- What works is E-E-A-T: original research, credible sourcing, third-party advocacy, and branded mentions across trusted platforms. Keyword stuffing does not work, and can actively hurt visibility.
- Branded mentions, not backlinks, are the strongest predictor of AI visibility, according to Ahrefs’ study of 75,000 brands.
- Watch for red flags: guarantees, urgency-driven acronym stacking (GEO, AEO, AIO, LLMO), claimed “insider access” to AI models, and prompt-injection tactics now formally classed as spam by Google and Microsoft.
- AI visibility tools are useful for direction, not certainty. None of them see real user prompts or have an official data feed from any AI platform.
- This has happened before. The current GEO gold rush mirrors the early-2000s SEO gold rush, right down to the guaranteed-rankings pitch and the vendor who stops answering the phone.
I think I've seen this film before
Previous iterations of this promise focused on guaranteed first-page Google rankings, delivered via some secret insight into how ranking worked. The results fairly rarely materialised. Some vendors hid keywords as white text on white backgrounds. Others built link farms that worked well, right up until Google noticed, at which point the client’s site disappeared from the index and the vendor stopped answering the phone. It took the industry the best part of a decade to live that era down, and it undermined a huge amount of creativity, technical skill and hard work by talented SEO and link-building PR professionals.
Generative engine optimization, or GEO, is the craft of earning visibility in AI-generated answers. It’s real, and it matters. It has also attracted the most enthusiastic crowd of chancers the industry has seen since that first SEO gold rush. Anyone holding budget for AI search visibility right now needs a working nose for the difference between legitimate practice and expensive theatre.
The gold rush is real
Let’s be fair to the hype before we knock it: consumers genuinely are changing how they search. SparkToro’s analysis of Similarweb data this year found over two-thirds of US Google searches ending without a click to any website. Similarweb has tracked AI referral traffic to websites more than tripling in a year. Gartner’s much-argued-over prediction that traditional search volume would fall 25% by 2026 hasn’t aged perfectly, but the direction was on point. And the traffic that does arrive from AI answers seems to be unusually valuable; SEMRush found AI search visitors converting at several times the rate of ordinary organic ones.
Budgets have, of course, moved accordingly. Conductor’s research found 94% of digital marketing leaders planning to increase GEO spend in 2026. So, there is real money, moving fast, into a channel very few people properly understand yet.
That combination of genuine shift, mobile budgets and thin understanding is the natural habitat of the snake oil salesman. Nobody sold miracle tonic in towns of yesteryear where everyone understood medicine. One case reported last year involved a mid-sized e-commerce firm paying $50,000 to a self-described GEO expert who’d promised guaranteed placements in AI answers. Six months on they had precisely nothing.
The single most useful thing to understand about AI search
Large language models are probabilistic – you can ask ChatGPT, Claude or Gemini the same question twice and you can get two different answers citing two different sets of brands. Not because anything is broken, that’s simply what the technology is.
The scale of the variability has actually been measured. Profound, one of the more credible AI visibility tools, re-ran around 80,000 identical prompts a month apart and found that 40 to 60% of the domains cited in the answers had changed completely in that month. Not reshuffled, gone completely, replaced by others. Stretch the gap to six months and the churn reaches 70 to 90%.
Sit with that for a second, because it quietly demolishes half the sales pitches in this market. If the answers move that much for identical questions asked by the same tool, a “guaranteed citation” isn’t ambitious, it’s deeply disingenuous and commercially dishonest. Add personalisation, chat memory, location, and the fact that every engine retrieves sources differently, and there is no such thing as a placement anyone can promise you.
The illusion or promise of some form of “special access” or “insider knowledge” is, also, utterly false. Google has said in terms that third parties claiming insight into its AI systems are guessing. OpenAI operates no submission service; there’s no index to be pushed up, no directory to be listed in.
The 80/20 litmus test for GEO vendors
So how do you sort real practitioners from the rest? The best filter we’ve come across is one used by Jeremy Moser, who runs an agency in this space: if a GEO provider doesn’t tell you upfront that roughly 80% of AI visibility is plain, fundamental SEO, they’re selling snake oil.
The 80% isn’t a platitude, either; it’s roughly what the evidence supports. The academic paper that coined the term generative engine optimisation tested which tactics genuinely improved a site’s surfacing in AI answers.
What worked was almost comically virtuous and music to the ears of PR professionals and earned media specialists worldwide. Quoting credible people, citing sources, real statistics, original research and third party advocacy from trusted news outlets. In contrast, what didn’t work was keyword stuffing, which produced little to no improvement and sometimes made visibility worse. The tactic that wrecked the early web simply doesn’t function on language models, and there’s something deeply satisfying about that.
The other big finding concerns what happens off your own website. Ahrefs studied 75,000 brands and found the strongest predictor of AI visibility wasn’t backlinks, the old currency, but branded mentions across the wider web, by a wide margin. And when researchers pull apart millions of AI citations to see where the engines actually look, the same names come up again and again: Reddit, Wikipedia, YouTube, Facebook, LinkedIn, editorial press, review platforms. Places where reputations are earned rather than bought, in other words. E-E-A-T, Google’s quality-rater shorthand for experience, expertise, authoritativeness and trust, has quietly become the operating logic of LLM visibility as well.
If you’ve spent your career in PR, you’re allowed a small smile at this point. Earned coverage and credible third-party mentions, the stuff parts of the performance-marketing world spent years dismissing as unmeasurable, turn out to be the raw material AI answers are made from.
GEO red flags
Which leaves the other 20%: structured content, clean technical foundations, being present and accurate in the sources the engines lean on. All real work, all worth doing. The trouble is the thicket of nonsense that has grown up around it is dense and can be misleading.
Guarantees, first and always are the ultimate red flag. Guaranteed citations, guaranteed placements, a promised uplift with a percentage attached. You now know why that’s impossible; treat it as you’d treat a hedge fund manager guaranteeing returns or Thomas Tuchel promising that it’s coming home.
The second major watch out is urgency plus acronyms. GEO, AEO, AIO, LLMO, each sold as a separate discipline, you’re already dangerously behind on. John Mueller at Google put it nicely: the higher the urgency, and the harder the push of new acronyms, the more likely you’re looking at spam and scams.
The third is some sort of magic ingredient – a proprietary “LLM submission” technology, claimed relationships with AI platforms, insider knowledge of the models.
Moreover, many are dressing up algorithmic manipulation as innovation. Prompt-injection buttons that instruct AI assistants to recommend the brand and comparison pages generated by the thousand, as well as listicle farms declaring their client number one in everything.
Microsoft and Google have both now formally classified the prompt-injection trick as spam, and the penalty waves have started rolling through, exactly as they did for link farms twenty years ago. The old pattern completes itself: the tactic works until it abruptly doesn’t, and it’s the client who wears the consequences.
And false precision. A dashboard telling you your AI visibility score is 34, and that the vendor’s work will lift it to 41, is measuring noise to two significant figures.
What AI visibility tools can and can't tell you
The legitimate ones are useful, with a caveat the good ones volunteer themselves. Platforms like Profound, Peec, Otterly and Evertune, along with what SEMRush and Ahrefs have built, all work the same basic way: fire large volumes of prompts at the engines and track how often a brand appears. Done properly, that gives you directional insight. Are we broadly visible or invisible? How do we compare with competitors? Which sources do the engines cite in our category? Is the line moving over quarters?
What none of them can give you is certainty, because the underlying system doesn’t contain any. No tool sees real user prompts. None has an official data feed from any platform. They are all photographing something that is constantly moving and the photograph blurs. The honest vendors say this unprompted, and that’s really the meta-test running underneath this whole piece: in AI search, the legitimate operators lead with the uncertainty.
What this means for brands
What actually works in AI search visibility is what would have worked in 2002, minus the shortcuts. Ignore anyone selling certainty about a probabilistic system. Put the budget into things with evidence behind them: well-sourced content with genuine expertise in it, earned media coverage, credible mentions in the places both people and machines have learned to trust. Measure directionally, over quarters, and hold precise scores lightly. And when the email promising guaranteed ChatGPT or Claude citations lands, remember how the last version of that story ended.
FAQs
- What percentage of GEO is just SEO? Roughly 80%, according to the litmus test used across the industry and supported by the academic research that coined the term generative engine optimization. Structured content, technical foundations and source accuracy make up the remaining 20%.
- What actually improves AI search visibility? Original research, real statistics, credible sourcing, and earned third-party advocacy from trusted outlets. Branded mentions across platforms like Reddit, Wikipedia, YouTube and editorial press are the strongest known predictor, ahead of traditional backlinks.
- Are AI visibility tools like Profound or Peec worth using? Yes, for directional insight over quarters, such as category comparisons and source tracking. They can’t offer certainty, since none of them have access to real user prompts or an official platform data feed.
About the author
Simarin Tandon | Junior Digital Account Director
Having worked with brands across the Beauty & Wellness, FMCG, FinTech, and Home & Lifestyle sectors, Simarin focuses on driving acquisition and growth, whilst managing the Digital team at brandnation.
A curious marketer, Simarin’s finger is always on the pulse when it comes to performance and digital updates across both paid and organic platforms.
