Someone read ChatGPT's network traffic, here's the tea
This week's a bit of a double bill - one half is how the machines actually work, from ChatGPT's source-picking logic laid bare in its own network traffic to a properly reusable way to run content audits in Claude. The other half is a reminder that behind every AI Overview and onboarding flow, there's still a slightly anxious human trying to make a decision. Here are your five.
1. Stop trying to prove the ROI of GEO - ask this instead
Foundation's latest deep dive tackles the question every client eventually asks: what's the actual ROI on GEO work? The honest answer, according to VP of Strategy James Scherer, is that you mostly can't calculate it - the moment a brand gets mentioned in ChatGPT, a Reddit thread or an AI Overview happens somewhere you don't own and can't track, so attribution just breaks. The suggested reframe: treat GEO like PR or brand awareness spend rather than performance marketing, and ask what the opportunity cost is of letting competitors own the AI answers in a category while everyone waits for a tidy spreadsheet that isn't coming. One case study saw a client's mention rate across the 100 most-cited Reddit threads in their category climb from 12% to 73% in three months - a number worth putting in a slide even without a conversion rate attached. Foundation
2. What ChatGPT's own network traffic reveals about how it picks sources
SEO pro Suganthan Mohanadasan spent two days digging through ChatGPT's raw network traffic - not the answers, the JSON underneath - and found the label it stamps on every source it pulls: result_source. Two of its four values are commercial web scrapers doing most of the fetching for anything outside a small allowlist of licensed publishers. The worrying part for clients with JavaScript-rendered pricing tables: in its own saved reasoning, the model repeatedly tried to read a brand's pricing page, found the numbers loaded behind JavaScript, gave up, and cited a review site instead - the client's own facts, sourced from somebody else. Worth an audit pass on any client site where pricing, specs or key numbers sit behind JS rendering, toggles or images. Suganthan.com
3. Turn one-off Claude prompts into an actual content audit workflow
Search Engine Land's Tania Brown lays out six content audit workflows worth building as reusable Claude skills rather than one-off prompts: topical gap analysis, freshness checks, brand voice consistency, AI retrievability, coverage comparison and performance triage. It's the difference between asking Claude to "check this page" every time a client emails, and building something you can point at any client's content library for a consistent, repeatable read - the kind of consistency that matters when you're running audits across dozens of sites. Worth starting small: pick one client, build a single workflow (freshness is the easiest place to begin), then stack the rest on top. Search Engine Land
4. Your B2B clients' customers aren't as rational as they think they are
A short one from Frontera worth sitting with: the idea that B2B buyers are cold, logical decision-makers while consumers are emotional impulse-buyers doesn't really hold up - same humans, higher stakes if anything, managing career risk or just tired of hearing the same complaint in every meeting. Consultancies and B2B clients fall into this trap constantly, with websites that talk about their own vision, services and awards and nothing about the buyer's actual anxieties. Worth revisiting with any B2B client's messaging: ask what emotional job the purchase is doing for the buyer, not just what problem it solves on paper. Frontera
5. The onboarding moment that actually keeps users around isn't the one you think
ProductLed draws a useful distinction between a product's "Aha moment" - the emotional realisation it's going to be valuable - and "activation", the first time a user actually experiences that value hands-on. Mixing the two up is a common reason onboarding flows underperform, since different user segments hit their Aha moment at different points and for different reasons - a single generic onboarding flow leaves retention on the table. Worth a conversation with any client complaining about early drop-off: ask what their Aha moment actually is, by segment, and whether onboarding is built to get users there fast. ProductLed
Small business shout-out
This week's shout-out goes to Tin Can, a Wi-Fi landline shaped like a tin can for kids too young for a smartphone but old enough to want to call their friends - no apps, no texting, no strangers, just voice calls to parent-approved contacts with quiet hours built in. Know a small business we should feature? Reply and let us know.
