Generative engine optimisation (GEO) is how brands earn citations in AI answers. Our traffic fell 50% after AI Overviews launched. Here is what worked.
In late 2024 I watched half of our website traffic disappear. Rankings hadn’t moved. We still sat on page one for around 75 of our target terms. The clicks had simply stopped arriving, and our HubSpot dashboard looked like someone had switched the internet off.
I made a TikTok about it at the time and nobody seemed to be talking about the problem, which worried me more than the traffic did. Eighteen months on, that drop has turned into the biggest shift in search since Google launched in 1998. This post is the written version of our sold-out GEO webinar, with the data we shared on screen.
Why has your traffic dropped when your rankings haven’t?
Because AI Overviews answer the question on the results page, so even number one rankings lose most of their clicks.
The publishers reported it first. DMG Media told the Competition and Markets Authority that click-through rates on desktop fall from 25.2% to 2.8% when an AI Overview appears, an 89% drop. The Mail’s own study of 5,000 keywords found clicks fell 56% on desktop and 48% on mobile even where it ranked first. Business Insider lost 55% of its organic search traffic between 2022 and 2025, then cut 21% of its staff.
Semrush’s research puts around 60% of Google searches ending without a click on anything, and that rises above 80% when an AI Overview is present. Reach, the owner of the Mirror and the Express, announced 321 job cuts in its biggest ever restructure while the NUJ asked how AI featured in it. Someone always pays for vanished traffic. In journalism, it was paid with jobs.
What is generative engine optimisation (GEO)?
GEO is the practice of making your brand visible inside AI-generated answers. Authority earns inclusion. Keywords on their own now earn very little.
SEO was built on links and keywords. GEO works differently because tools like ChatGPT, Gemini and Google’s AI Mode compose one answer from sources they trust, then cite the brands those sources mention. It blends PR and owned content with technical readiness. Welcome back, blogging. We’ve missed you.
If your brand isn’t cited, you have no control over how AI describes you. I learned this the embarrassing way when an AI transcript of our own podcast invented a guest’s company name. Quite ironic for a show called Embracing Marketing Mistakes.
What actually happened to our traffic and leads?
Our sessions dropped nearly 50%, but lead quality went up and leads from ChatGPT rose 500%. Fewer visitors, better ones.
Our HubSpot data tells the story properly. Sessions fell off a cliff in the weeks after AI Overviews rolled out, yet key events such as contact form fills held up and then climbed. AI referral traffic went from 44 visits in January 2025 to around 200 a month by the autumn. We have won clients who told us not to bother pitching because ChatGPT had already shown them our work.
“Traffic has gone down with a bang, but actually the quality of traffic that we’re getting is better.” Chris Norton
Organic search leads fell 55% over the same period. PPC is propping plenty of businesses up, but cost per acquisition is rising because everyone who lost organic visibility is bidding on the same terms. That maths gets worse every quarter.
How does AI decide which brands to trust?
Through signals: cues of credibility, consistency and verifiability across the web. Editorial media carries the most weight and social media almost none.
Research by Hard Numbers analysed which sources shaped GPT-4’s answers about the reputation of Forbes’ top 100 brands. Editorial media drove 61% of the content, owned media 44%, and social media less than 1%. When the question was ‘is this company trustworthy?’, editorial coverage influenced 65% of answers.
“AI doesn’t rank links. It chooses sources that it trusts. The stronger your signals, the more inclusion you’re likely to have.” Will Ockenden
That is why PR has quietly become a search discipline. A journalist writing about you is third-party proof. A wall of unattributed website copy is just noise the model can generate itself.
How do you get your brand cited in AI answers?
Structure your content so AI can extract it, build earned authority off your site, and remove the technical barriers that stop models crawling you.
This is the checklist we now run for clients, and it starts with an audit of where you appear today across ChatGPT, Gemini, Perplexity and Claude. Every brand’s citation gap is different.
- Structure pages with clear headings, FAQs and schema markup so answers are easy to lift
- Open each page with a short, quotable answer rather than a slow build-up
- Show your experts: names, biographies, credentials and quotable lines
- Earn editorial coverage in credible trade and national media
- Secure directory listings and keep citations of your brand consistent everywhere
- Run a reviews programme, because ratings demonstrate trustworthiness to the models
- Add an llms.txt file, fix broken links and keep key content out from behind logins
- Refresh facts regularly, since stale claims get skipped or, worse, hallucinated
Listen to the full episode
Watch it on YouTube or listen on your podcast app via embracingmarketingmistakes.co.uk. If you want to know where your brand stands, you can book a GEO audit call with me.
Embracing Marketing Mistakes is the weekly podcast hosted by me as the founder and my lovely business partner Will Ockenden, where senior marketers share the screw-ups that taught them the most. Over 150 episodes and 150,000 downloads so far, with surprisingly few legal letters.
Here is the question worth sitting with. When a prospect asks ChatGPT who the best provider in your market is, does your brand come up? Because somebody’s does.
About The Author
