Why 67% of Brands Fail at GEO: Avoid These Mistakes

Your product is genuinely good. Your website looks the part. And yet when someone asks ChatGPT or Google's AI Overviews about your category, you're nowhere. The AI cheerfully recommends three competitors and skips you entirely.

That stings. We see it every week. A brand with real expertise gets quietly written out of the conversation because nobody told the machines it exists.

So let's talk about generative engine optimisation, what the data actually shows works, and why so many brands keep getting GEO wrong even when they're trying hard. We run these campaigns for a living, so this is what we see in the wild, not theory borrowed from someone else's blog.

What is generative engine optimisation?

Generative engine optimisation is the practice of getting your content cited, quoted and recommended inside AI-generated answers from tools like ChatGPT, Perplexity, Google's AI Overviews and Gemini. Where classic SEO chases a blue link on a results page, GEO chases a mention inside the answer itself.

That difference matters more than it sounds. The answer is increasingly where the journey ends.

Pew Research Center tracked the real browsing behaviour of 900 US adults across nearly 69,000 Google searches in March 2025. When an AI summary appeared, people clicked a traditional link in just 8% of visits, compared with 15% when no summary showed up, according to Pew Research Center's study on AI summaries and clicks. Roughly half the clicks, gone, the moment a summary turns up.

This is not a future problem dressed up as a trend. Rand Fishkin's analysis of Datos clickstream data found that 58.5% of US Google searches already ended without a single click to the open web, per the 2024 SparkToro zero-click search study. AI answers are accelerating a behaviour that was already well established. Being the cited source is no longer a nice-to-have, it's how you stay visible at all.

Why most brands fail at GEO (the honest list)

Here's the uncomfortable bit. Most GEO failures aren't caused by some exotic algorithm. They're caused by brands doing sensible-sounding things that don't actually move the needle. These are the patterns we see most often.

1. Chasing magic files instead of writing better content

The single biggest myth doing the rounds is that you need a special AI file or secret schema to get into AI answers. People burn weeks on this.

Google says the opposite, in plain English. Its own documentation states that "you don't need to create new machine readable files, AI text files, or markup to appear in these features" and that "there are no additional requirements to appear in AI Overviews or AI Mode," per Google Search Central's guidance on AI features. The same foundational SEO that earns you rankings earns you citations.

Our take: don't ignore llms.txt entirely, just keep it in perspective. We dug into whether it actually drives citations in our breakdown of what the llms.txt data says about AI citations, and the honest answer is it's a low-effort housekeeping item, not the thing that gets you quoted.

2. Burying the answer three paragraphs deep

AI models extract the part of your page that most cleanly answers the question. If your sharpest sentence is hiding under a long warm-up, the model grabs a competitor's tidier line instead.

We've watched this happen on pages that ranked perfectly well in classic search. They had the right information. They just made the model work too hard to find it. Front-load the direct answer, then expand underneath. Position genuinely matters here, which is why we lean on the findings in our analysis of how ChatGPT pulls citations from the first 500 words.

3. No proof, no citation

This is the one with hard evidence behind it. The Princeton-led GEO study, presented at KDD 2024 by Aggarwal and colleagues, tested content tweaks across commercial generative engines and found that simple GEO methods can boost visibility by up to 40% in AI responses, according to the GEO: Generative Engine Optimization paper. The methods that worked best were adding relevant statistics, quotations and citations to source material.

Read that again. Adding statistics and cited sources beat stuffing keywords. The machines reward content that looks like it did its homework.

So the brands publishing thin, opinion-only "ultimate guides" with zero data are quietly opting out of the citation game. If you want the mechanics, we lay them out in our piece on how to get cited in ChatGPT and AI Overviews.

4. Treating every engine like it's the same engine

ChatGPT, Perplexity, Gemini and Google's AI Overviews pull from different places and weight sources differently. A page that gets quoted constantly in one can be invisible in another.

We've stopped pretending there's one universal recipe. The smarter move is to map where your category's answers actually come from, then optimise for those specific surfaces. Our broader framework for getting your brand into AI answers across engines walks through how we approach each one differently.

5. Being a blank space in the knowledge graph

If a model can't confidently work out what your brand is, who it serves and how it connects to the rest of your industry, it leaves you out. Not out of malice. Out of caution.

This is where weak entity signals quietly kill visibility. Inconsistent business details, a thin or missing presence on third-party reference sites, and content that never clearly states what you do. We go deep on fixing this in our guide to knowledge graphs and entity optimisation for AI search, and it's usually the least glamorous, highest-impact work in the whole project.

6. Publishing nothing worth citing

If your content only repeats what's already on page one, a model has no reason to cite you over the original. You become furniture.

Original data is the cheat code here. A small survey, a benchmark, a year of your own anonymised numbers, all of it gives engines something they can only get from you. We've seen one decent data study earn citations for over a year, which is exactly why we keep banging on about using original research to win AI search visibility.

What actually works: GEO best practices we'd stake our name on

Enough about failure. Here's the playbook we run, in roughly the order we run it. None of it is exotic. The discipline is in doing all of it consistently rather than cherry-picking the fun bits.

1. Lead with the answer. Put a clear, self-contained answer in the first couple of sentences under each heading. Write it so it makes sense lifted out of context, because that's exactly what an AI will do to it.
2. Add proof to every claim. Statistics, dates, named sources, expert quotes. The Princeton data is clear that this is what triggers citations, so treat unsupported assertions as a liability.
3. Fix your entity footprint. Consistent name, address and details everywhere, a tidy Organisation schema, and clear "we are X, we help Y do Z" statements. Give the machines no room to misunderstand you.
4. Build topical depth, not just one big page. A strong hub page plus genuinely useful supporting articles, linked sensibly between each other, signals real authority on a subject.
5. Publish something only you can publish. Original research, proprietary data, a contrarian-but-defensible point of view. Become the primary source, not the echo.
6. Keep earning real authority signals. Mentions, citations and links from credible sites still feed how confidently engines describe and recommend you. Solid technical and content SEO remains the foundation the whole thing sits on.

Notice what's missing from that list. No tricks. No gaming. GEO done properly is mostly just SEO held to a higher standard of clarity and evidence.

Does GEO actually drive business, or is it a vanity metric?

Fair question, and one we get from finance teams constantly. Citations feel fluffy until you connect them to revenue.

The retail numbers are doing that connecting fast. Adobe Analytics reported that traffic to US retail sites from generative AI sources jumped 1,200% between July 2024 and February 2025, and that AI-referred visitors browsed 12% more pages with a 23% lower bounce rate than other sources, per Adobe's analysis of generative AI retail traffic. The volume is still small relative to classic search, but it's higher-intent and growing at a rate you can't ignore.

Our read: GEO traffic today is a bit like organic search traffic in 2009. Small, underpriced and about to matter a great deal. The brands building citation authority now will be the defaults their competitors are still trying to dislodge in 2027.

How to measure GEO without losing your mind

You can't improve what you don't track, but GEO measurement is messier than opening Search Console. Here's the lean version we use.

Build a citation watchlist. Pick 20 to 30 questions where you genuinely should appear, then run them across ChatGPT, Perplexity, Gemini and AI Overviews on a regular cadence. Log who gets cited, in what context, and whether you're framed as the expert or an also-ran. Yes, it's a faff. Do it anyway, or use one of the emerging tracking tools to take the grunt work off you.

Separate branded from unbranded. Getting cited when someone names you is table stakes. Getting cited for the broad category question, when nobody asked about you specifically, is the prize. That's market share inside the answer.

Watch referral and assisted conversions. Some AI platforms now pass referral traffic, so tag it and watch what it does. More often, AI is the first touch and a direct visit closes the deal later, so don't judge GEO on last-click alone or you'll wildly undervalue it.

When the AI gets you wrong

Here's a failure mode nobody warns you about. Sometimes the problem isn't invisibility, it's an AI confidently telling thousands of people something false about you.

We've seen models invent pricing, misattribute services, and merge two companies into one. It's maddening, and the fix is rarely a single feedback button.

The durable solution is to be the better source. Publish a crisp, well-structured, clearly-dated correction on your own site, make it unambiguous, and then earn legitimate references to it so engines have something more reliable to lean on. This is exactly the defensive groundwork we cover when we help brands shape how AI describes them, and it's worth getting ahead of before a hallucination costs you a sale.

So where should you start?

If you only do three things this quarter, do these: front-load the answers on your best pages, add real proof to every important claim, and clean up your entity signals so the machines stop guessing.

That alone puts you ahead of most of your category, because most of your category is still arguing about secret files and ignoring the boring fundamentals that actually earn citations.

The brands winning AI search in 2026 aren't the ones with the fattest budgets. They're the ones who treated it as a system early, diagnosed the real bottleneck, and fixed it methodically. That's the whole game, and it's the way we approach every AI search visibility project we take on.

If you'd rather not run the experiments yourself, tell us where you're getting beaten in AI answers and we'll show you exactly why, and what we'd fix first. No magic files required.