PPC.London held its inaugural event last Friday, run by Martin Svilenov of Averank, and it landed exactly where the industry's head is at right now. Not another AI hype panel, but practitioners in accounts every day sharing what's actually breaking and rebuilding in paid media. Here's what stood out, and why it matters for how we run PPC at Two Trees.
A theme that came up across several talks: ChatGPT Ads, and what they mean for how paid media gets planned. If you're a client reading this and wondering why your paid media plan needs to look different in 2026, this is why.
Igor Ivitskiy (Doctor Ads) gave one of the sharpest talks of the day: before you trust a strong ROAS number, count the sales behind it. His rule of thumb: with 25 sales, luck alone can move ROAS by 20% or more; with 100 sales, it's still 10% or more. Split the days odd vs even: if you get two different ROAS numbers, it's too early to call it a win.
He grouped the ways a good-looking number can mislead you into three buckets:
What this means for clients: this is exactly why we don't report on ROAS in isolation. We look at incrementality, not just the headline number, before we scale spend on the back of it.
Mery Hayles (Adthena) shared the first hard numbers on ChatGPT Ads: a $1B annualised revenue run rate as of August 2026, in under 200 days since OpenAI's advertising pilot launched, with 5,471 unique advertisers tracked in the UK and 20k+ in the US.
The honest takeaway: most advertisers aren't seeing direct ROI yet. Most prompts are informational, so ChatGPT is behaving as a top-of-funnel, visibility-building channel rather than a direct-response one right now. Adthena's own case study showed how quickly this can shift though: after auditing which brand-comparison prompts they were missing from and rebuilding tighter ad groups around them, their ad frequency on comparison prompts went from 0% to 22% within days. A prompt-led negative targeting feature has also launched in the US, though not yet in Europe.
What this means for clients: don't expect ChatGPT Ads to behave like a search campaign yet. Budget it for visibility and brand presence in AI answers, audit where competitors are showing up in comparison prompts where you're not, and revisit ROI expectations as the format matures.
Sophie Logan's talk on why B2B PPC "isn't working" resonated. The platforms weren't built for B2B: long sales cycles, high CPCs, niche products, and low conversion value per click all work against the standard playbook. Her fixes:
What this means for clients: if you're a B2B client wondering why your cost-per-lead looks worse than a B2C competitor's, it's not necessarily a targeting problem. It's often a measurement and handoff problem further down the funnel.
Azeem Ahmed's framing was one of the most useful mental models from the day: the funnel isn't fractured, it's forked. Different customers now default to different starting points, a Google habit, TikTok search, or an AI answer they never question, and Ofcom figures show 54% of UK adults now use AI tools, up from 31% in 2024. That shift in default behaviour is happening right now, not in some future roadmap.
It's worth flagging just how far along the UK is on this. We have the highest zero-click search rate of any country studied, ahead of the US, where 68%+ of Google searches already end without a click. Around 30% of UK Google searches now trigger an AI Overview. As Azeem put it, people aren't leaving Google. They're leaving the click. For a UK-focused agency, that's not a future risk to plan around, it's already the reality most of our clients' campaigns are running in.
He mapped four buyer types, each needing a different creative and proof approach:
Two of the biases behind this are worth flagging on their own: status quo bias (switching feels costly even when the alternative is clearly better) and default effect (we don't question the pre-selected option). Both explain a lot about why customers stick with a brand's default channel even when a competitor is objectively cheaper or better.
What this means for clients: a single-funnel, single-attribution-model view of paid media is now actively misleading. If your audiences, creative and attribution assume one path to purchase, and one built purely around the click, you're paying to reach people who increasingly aren't there. Visibility inside AI answers is becoming as important as the click itself, and that's not something PPC alone can fix. It needs to sit alongside SEO and AI search strategy.
Aleksei Rozhnov (Thred) shared hands-on learnings from testing ads inside LLM platforms. His framing: keywords are ice cubes, prompts are snowflakes. An AI query is a brief, not a keyword, closer to a broad-match campaign that actually makes sense because the model has real context. The page is no longer the atomic unit; the answer is. Formats built for the chat interface earn attention, not formats built for a landing page.
Three practical takeaways he gave:
The clearest signal from the day, coming from someone who's spent the last decade-plus mainly in SEO: success in search now genuinely requires a multi-channel approach. Paid media can't be run in isolation from SEO and AI search visibility. They're feeding the same buyer, who's increasingly forking across Google, AI answers, and platform-native search before they ever reach a landing page.
For our clients, that means:
This is the direction we're already building Two Trees' offering toward, and PPC.London was a useful validation that the rest of the industry is heading the same way.