What do we do about ChatGPT, Perplexity, and AI agents?


What do we do about ChatGPT, Perplexity, and AI agents?

Last week, I presented at BrightonSEO - the world's largest search marketing conference - where thousands of marketers and business leaders gathered to discuss what's actually working in digital marketing right now.

My session: The Agentic Era: How to Optimise for AI Agents, Not Just Humans [access slides here].

As you'd expect, AI dominated most conversations. But here's what surprised me:

Nobody was panicking.

Instead, conversations were practical: What's working. What's not. How to track it.

The search community isn't scrambling. They're adapting. Fast.

And I can't think of a better group to lead AI strategy than the people who understand both humans and machines.

If you're already doing SEO well, you're halfway there to figuring out your AI strategy. The other half? Giving your team time to test, track, learn, and adapt.



Meanwhile, elsewhere this week in search and AI...


What Public Companies Recently Said About AI and Search

Booking.com:
"Travel clicks that are coming to us from
traditional search are still going up year-over-year. The number of leads we're receiving from large language models is relatively small, but it is growing."

Translation: Traditional search still dominates. AI traffic is tiny but increasing.

Wayfair:
"We are actively shaping the future of agentic commerce with a clear dual-pronged strategy. First, we view this as a new additive channel for growth, so we will be where our customers are. This begins with ensuring a vast and specialised catalog is deeply integrated and accurately represented on leading AI and search platforms including Google, OpenAI and Perplexity."

Translation: AI agents are a new sales channel.

CoStar Group (Homes.com, Apartments.com):

"SEO remains the foundation of AEO and GEO, though a portal's brand, content, and context remain the key building blocks for success. Today, GEO is sub-1% of top-of-funnel acquisition... but will become a much bigger top-of-funnel traffic feed. Many believe that traffic from GEO may be monetised the way Google monetised SEO with SEM."

Translation: AI traffic is less than 1% today but growing. SEO skills are the foundation. This could become the next paid search.

The consensus: All three agree - AI traffic is tiny today (sub-1%), traditional search still dominates, but AI is growing consistently and SEO fundamentals remain the foundation.

👉 Public Company Commentary on SEO and AI in 2025 - Detailed.com

PayPal + OpenAI: ChatGPT Just Became a Checkout Page

Yesterday, PayPal announced a landmark deal with OpenAI: they're embedding payments directly into ChatGPT.

Starting in 2026, users will be able to discover products, compare options, and complete purchases - all without leaving the conversation. PayPal handles payments, fraud protection, tracking, and disputes.

Compare the funnels:

• Google Shopping: Search → click → website → cart → checkout

• Amazon: Search → product page → add to cart → checkout

• ChatGPT + PayPal: Ask → recommend → buy → done

The funnel just collapsed.

What this means: 800 million weekly ChatGPT users will be able to complete purchases without ever visiting your website. Product data quality, content, and popularity determine whether you're recommended. The time to optimise is now.

👉 OpenAI and PayPal Team Up to Power Instant Checkout and Agentic Commerce in ChatGPT - PayPal Newsroom

👉 PayPal signs deal with OpenAI to become the first payments wallet in ChatGPT - CNBC Finance

Your Analytics Just Broke: What ChatGPT's Atlas Browser Means

Last week, ChatGPT launched Atlas - a Chromium-based browser built directly into ChatGPT.

If you've been watching Google Analytics, you might have noticed something strange: a spike in "new users" that aren't actually new.

What's happening: When users switch from Chrome to Atlas, their GA cookies don't transfer (by design, for privacy). GA4 treats returning customers as brand new visitors.

The impact: New users inflated, returning users deflated, attribution messy.

Who's affected: Early adopters - marketers, developers, tech enthusiasts.

If you have unexplained new user spike - that's Atlas.

👉 Introducing ChatGPT Atlas - OpenAI

👉 When a New Browser Launches: What I'm Seeing in Google Analytics for ChatGPT’s Atlas - Analytics Playbook

How Does ChatGPT Traffic Perform Compared to Traditional Channels?

A new study analysed 973 e-commerce sites ($20B combined revenue) to see how ChatGPT referrals actually perform.

The results:

ChatGPT traffic: 0.2% of total (200x smaller than Google)

Conversion rates: lower than all traditional channels except paid social

Average order values are declining over time

Growth rate: 40% month-over-month

The consensus from the researchers: "This evidence challenges narratives of LLMs as immediate Google killers but also points to potential for long-term channel evolution."

Why this matters: Brands shouldn’t abandon traditional search just because LLMs are hot. But right now is the window to test while stakes are low and learning is cheap. By the time AI traffic is 5% of your total, early movers will have 12-24 months of optimisation advantage.

👉 ChatGPT Referrals to E-Commerce Websites: Do LLMs Outperform Traditional Channels? - University of Hamburg & Frankfurt School of Finance & Management [warning: 61 pages long PDF]



Living 48 Hours Without AI


A.J. Jacobs tried an experiment for The New York Times: live without AI for 48 hours. He couldn't:

Unlock his phone (facial recognition)

Turn on lights (energy grid uses machine learning)

Brush his teeth with tap water (NYC reservoir system uses AI sensors)

Take the subway (fare monitoring, electronic signs)

Pay with credit cards or use an ATM (fraud detection)

Order lunch via Resy (machine learning)

Walk his dog without anti-facial-recognition sunglasses (street cameras)

We won't spoil what happened, but what's clear is AI isn't future tech. It's already the invisible infrastructure we live inside.

👉 48 Hours Without A.I. - The New York Times


Michael Chidzey, Editor of The Signal

The web’s noisy. The Signal cuts through it.

I built my first website in ’98. And I never stopped. Growth became the craft. Craft became Good Signals.

At Good Signals, we help brands that deserve to do better online - do just that. From scaling startups and e-commerce stores, to managing complex websites and guiding billion-pound organisations, the mission’s always been the same: build growth that lasts, and makes business - and the web - better.


Last thing.

We hope you enjoy this email from Good Signals. However, if you no longer wish to receive any further communications from us, you can simply unsubscribe below.

Good Signals, Studio 1, The Bottle Factory, 12 Ossory Road, London, Greater London SE1 5AN

P.S. You can update your preferences here.

The Signal by Good Signals

Your website is like a dog. That dog needs love, care and attention. Go on, sign up to The Signal and discover how to whip up some Scooby Snacks for your prize pooch.

Read more from The Signal by Good Signals

Everything wrong with optimising for ChatGPT, Claude, AI Mode, Perplexity, AI Overviews… It's occurred to me over the last year that we've talked endlessly about AI - new models and rollouts, updates, features, growth numbers, even the personalities - but we've barely touched on the one question that keeps coming up behind the scenes: "How do you rank higher on LLMs?"ChatGPT? AI Mode? Perplexity? Claude? Should each be optimised differently?These are fair questions.But here's the thing: they...

Everybody Loves Ray-Bans (Hopes Mark Zuckerberg) From the search bar to your customer’s line of sightTen years ago, Google Glass flopped. Too early, too awkward, too gimmicky.This week, Mark Zuckerberg tried again. At Meta Connect, he unveiled the new Ray-Ban Display glasses. And the first reviews are… surprisingly positive.The pitch this time isn’t “cool tech on your face.” It’s AI. Meta is framing Ray-Bans as the gateway to personal superintelligence - an assistant that sees what you see,...

What are people really using ChatGPT for? If you thought ChatGPT was mostly for work - emails, code, reports - think again. The largest study of its kind, from OpenAI and Harvard economists, analysed 1.5 million conversations and found something very different: nearly three-quarters of all usage is now non-work. This shift has big implications. People are turning to AI not just as a productivity tool, but as a life advisor. The data shows three categories dominate: Practical Guidance, Seeking...