When AI Overviews rolled out, traffic dropped. Top-of-funnel disappeared overnight for a lot of sites. And almost everyone — clients, agencies, the LinkedIn commentariat — blamed the wrong thing.

The story went: "ChatGPT is eating Google's lunch." Easy narrative, mostly wrong. ChatGPT had been around for years before most people noticed it, and the share of search traffic going to LLMs sits at around 1%. It's incredible — that traffic converts roughly 20× better than the rest of your sources — but it's a rounding error on volume.

The actual culprit was Google itself. AIOs launched, informational searches stopped clicking through, and any site whose strategy leaned heavily on top-of-funnel content driving topical authority through long-tail clickthroughs to commercial clusters took the hit.

That's the bad news. Here's the good news: once you understand what actually happened, it's better than what you had before.

Clicks were never the goal

The whole industry got addicted to clicks because they were easy to count. But "make commercial pages rank for commercial intent" was always the real game — and AIOs barely touch that. They show up on informational queries. Buyers searching for "best [thing] for [use case]" still land on commercial pages. They're just doing more research before they get there.

That research now happens inside AI Overviews. The work shifted from "get the click" to "be in the answer."

If a long-form informational article used to bring 10,000 monthly clicks and now brings 3,000 plus a citation in the AIO, the brand exposure can actually be net up. Same brand, more eyeballs, none of the bounce. It just doesn't show up in the channels you used to track.

The customer arriving on your site is dramatically better

This is the part nobody talks about enough.

A pre-AIO SaaS funnel looked like: prospect lands on a comparison page, half-reads it, leaves, comes back, signs up for a walkthrough, takes the walkthrough, eventually buys. The walkthrough existed because the prospect didn't really know what they wanted yet.

Post-AIO, prospects arrive with the research already done. They've spent twenty minutes with an LLM working through what they need, what they should pay, what the trade-offs are. They land on the site and buy. They skip the walkthrough.

There's a measurable knock-on. Direct brand search has lifted across a number of clients — people research in an LLM, the brand sticks, they come back via Google to convert. That's a metric worth watching now in a way it wasn't two years ago.

What actually gets you cited

The word of last year was chunking. It still is. Restructure content into discrete, intent-matched chunks an AI can pull cleanly. We've been doing it for years, which is why we didn't take a beating when AIOs hit. Plenty of agencies haven't started.

Two specific moves that work:

1. Audit what you're being cited for vs. what you're not

A SaaS client of ours was being cited well for email but invisible for conversion tracking. The fix wasn't more content — it was one new chunk specifically about why they're good at conversion tracking, written and structured to be pulled. AIOs picked it up fast. Faster than classic SEO, actually. The AI-driven side of Google's algorithm moves quicker than the old one.

2. Find the angle the AI hasn't already written

Quality is everything, and it's getting harder. Original data, original framing, original angles. If your content is a recombination of what's already in the index, the AI has no reason to cite you over the source.

The case that flipped in six weeks

We had a client — physical product, pregnancy space — who came to us with a serious problem. The AI had decided their product was overpriced, not worth it, vaguely industrial. It was actively recommending a specific competitor. They couldn't sell anything, because the moment a prospect saw an ad and went to verify, the AI talked them out of it.

The strategy was one piece of long-form, deeply researched content built around the safety case for their product and the safety problems with the competitor's. We pulled in Reddit threads as a data source, added video, got five hard-won AU links from genuinely relevant publishers, and a small amount of PR.

Six weeks later the conversation had flipped across GPT, Gemini, and AIOs. The AI now actively steers prospects away from the competitor's product. The client placed two new container orders because stock was running off the shelf.

One article. Five links. One sentiment flip. Direct correlation to revenue.

That's the new game. You're not trying to rank a page — you're trying to change what an AI says when someone asks about you.

How to audit what AI is saying about you right now

Before you do anything else: run the audit.

Search the questions your customers actually ask, on every model that matters. Switch your VPN — Sydney, Brisbane, the markets you sell into. Different IPs return different answers. Recreate every search a few times to make sure you're seeing a stable picture, not a one-off. Screenshot everything.

You're looking for three things: where you're already cited, where the conversation is happening without you, and where the AI is saying something inaccurate or unfavourable. That third bucket is where the fastest wins live.

Then run the same audit after the work to benchmark the change. Tracking AI visibility is genuinely harder than tracking rankings — every query is its own thing, and the off-the-shelf tools haven't caught up. We use Hall (Sydney-based, building fast, the kind of partner you can give feedback to and see it land in the product), and we've built our own internal tracking on top. Expect this layer of the stack to mature significantly in the next twelve months.

"GEO vs SEO vs AIO" is the wrong argument

A lot of agencies are selling these as separate products. They're not.

What we do isn't search engine optimisation anymore. It's search everywhere optimisation. RAG doesn't care which platform the query started on — it pulls from wherever it can find authoritative material. AIOs use Google as one core data set. LLMs use Common Crawl. Perplexity blends live retrieval. The throughline is visibility across the open web — content, video, social, PR, mentions.

If your competitor is mentioned 1,000 times across the surfaces an AI considers authoritative and you're mentioned twice, you don't compete. The old 10× rule still applies — it just runs on mentions and citations now instead of backlinks.

A specific tactical tip: get visible in Common Crawl

GPT and most foundation models train on Common Crawl. They refresh their core training data every six to twelve months. If you want to be in the model itself — not just retrieved by it — you need to be visible in Common Crawl when those refreshes happen. Check your presence there. It's underrated and most teams ignore it entirely.

Where this goes

I was at a conference last week. Some people in the room were calling three years until the internet is unrecognisable from now. A few were calling twelve months. Either way, you need to be ahead of the curve, because by the time the change is obvious it's too late to position for it.

A few predictions worth holding lightly:

  • LLM growth is stabilising. The pace of capability change is still fast, but user behaviour is settling. Personalisation is the next big shift — models will increasingly search inside the context they already have on you.
  • Agentic is coming, but slower than the hype suggests. People still don't trust AI to do things on their behalf. ChatGPT walked back in-app bookings; Expedia's share price went up the same week. Trust is a real bottleneck. The first agentic use cases that win will be the ones where the friction was always too high — chasing $100 of insurance savings through a four-day form, that sort of thing.
  • Ad-supported AI search is also coming. Perplexity ran a Super Bowl spot kicking ChatGPT for it. Doesn't matter — it'll happen anyway, and the platforms that hold user trust through that transition will be the ones that win.

The thing nobody is saying

Most agencies pitching "GEO services" are repackaging what they were already selling. Most clients buying it are buying because they're scared, not because they have a strategy.

The right starting question isn't "how do we rank in AI?" It's "what's the actual business problem?" If the answer is "we can't sell because the AI is talking us down" — that's a content and authority problem, and it's solvable in weeks. If the answer is "we want to be everywhere a customer might search" — that's a long-term visibility play, and it's solvable in quarters.

The fundamentals haven't changed. Quality content, real authority, deep understanding of the customer, distribution everywhere they might look. The surfaces have changed, the tracking has changed, the speed has changed. The work is the same work. Just done with a different lens.

Adapt now. The window where being early matters is shorter than most people think.

Want a read on where you stand right now? Request a free Truth Gap audit — we probe the AI models with your buyers' deal-breaker questions and show you exactly what's missing.