Your audience may have left Google before you noticed

Rand Fishkin published SparkToro's developer audience data this morning, and one number deserves more attention than it'll get: 21% of US software developers at startups use Claude, an over-index of nearly 299% versus the general population. ChatGPT use among this group sits *below* the national average. Google use sits 14% below.

That last figure is the one to sit with. Developers — people whose entire professional life happens on the internet — use Google less than the average American. They've already routed around it.

The instinct in the SEO industry will be to treat this as a developer-marketing story. A niche finding about a famously contrarian audience. File under "developers are weird, here's what to do about it."

That reading misses the actual point. The developer audience isn't an outlier — it's a leading indicator. And if you're allocating SEO spend across B2B segments without asking which of your audiences have already moved, you're optimising for a distribution channel some of your highest-value buyers have quietly stopped using.

The audience is the variable nobody's controlling for

Most SEO strategy still treats Google share of search as a constant. The job is to rank, the venue is assumed, the only question is execution. Everything from keyword research to content calendars to budget allocation runs on the premise that your audience is on Google in roughly the same proportions as everyone else.

The developer audience isn't an outlier — it's a leading indicator.

Fishkin's data should puncture that. Different professional audiences use search engines at materially different rates. Developers under-index on Google by 14% and over-index on Claude by nearly 300%. SparkToro's earlier RevOps data showed a similar Claude over-index for that segment. There's a pattern: technical buyers — the people writing the cheques on a lot of B2B SaaS — have moved fastest and furthest.

This isn't the same as "AI search is growing." That's the industry framing and it's too soft. The harder version is: *for specific high-value segments, Google has already lost meaningful share, and you can identify which segments by job function*.

The audience is the variable nobody's controlling for.

If you sell to developers, your SEO programme is competing for a share of search that's already 14% smaller than the playbook assumes — and the bit you can't see, the Claude conversations, the Perplexity queries, the GitHub README that gets pasted into a chat — that's where buying decisions are forming.

Why this matters more in the UK than people think

UK B2B has a particular problem here. The default agency response to AI search uncertainty has been to wait — wait for measurement to mature, wait for share to stabilise, wait for the AI Overviews picture to clarify. There's a respectable version of that argument. I've made parts of it myself.

But "wait" assumes the audience hasn't moved yet. For broad consumer segments and most local-service businesses, that's probably true. For UK SaaS, fintech, devtools, agencies selling into tech-forward buyers — it isn't. Your buyer profile likely mirrors Fishkin's US sample more closely than it mirrors the average UK Google user. The numbers won't be identical, but the directional pattern almost certainly is.

Which means the businesses most likely to be reading SparkToro data and thinking "interesting" are the same ones whose own audiences have moved fastest. The diagnostic is in the mirror.

What "marketing to Claude" actually means

Fishkin's framing — *when a developer asks Claude about your category, do you show up?* — is the right question, but it gets misread immediately. People hear "marketing to Claude" and reach for GEO tactics. Schema. Conversational FAQ blocks. Whatever the vendor pitched them last week.

That's not what it means. Claude, like every other model, cites brands it's heard of. Its training data is the open web, weighted toward sources with editorial authority and earned references. The way you show up in Claude is the same way you show up in any serious citation surface: real publications mention you, your name appears in contexts the model treats as authoritative, your category language matches what your customers actually say.

In other words: it's brand. It's PR. It's the long boring work of being referenced by people who weren't paid to reference you.

The schema and structured data layer matters at the margin — EntityMap and schema can't fix what's actually broken in AI search, but they help retrieval. The generation layer, which is where citations actually happen, is downstream of brand. There's no shortcut to it. There's certainly not a tool you can buy that will fix it.

The measurement problem is worse for technical audiences

Here's the bit that should worry anyone allocating budget honestly. The AEO study released this week on arXiv looked at ChatGPT referral traffic to a single high-traffic domain. The headline finding: raw referral growth from ChatGPT was 5.7x, but once you control for platform-level growth using an untreated subset of the same site as a control, the actual treatment effect was 1.82x. And even that was statistically suggestive rather than conclusive.

Translation: most of what looks like AEO success is the platform getting bigger, not your work paying off.

Now apply that to a technical audience. Your developer-targeted content might be getting cited in Claude conversations at meaningful rates. You won't see most of it in Search Console — Search Console's new AI reports show impressions only, no clicks or queries. You won't see it in Claude analytics because Claude doesn't have an analytics dashboard for you. You'll see a vague rise in direct traffic, a few branded searches, the occasional Slack message saying "Claude recommended you."

For developer audiences specifically, the AI search measurement problem is a definition problem, and the gap is widest exactly where your buyers are. The tools assume Google is the centre of gravity. For your audience, it isn't.

What this means for how you spend

If you're running SEO for a business whose buyers skew technical, three things follow.

First, segment your audience before you segment your tactics. Most SEO briefs jump straight to keyword research without asking which channels the buyer actually uses. For technical audiences, that order is wrong. Channel allocation comes first; keyword work is a subset of one channel.

Second, the GitHub finding from Fishkin matters more than the Claude one in some ways. Developers over-index on GitHub by 64%. You can't buy ads on a pull request, but you can be the company whose open-source tooling shows up in the repositories developers actually use. That's not an SEO budget line, but it's a discovery investment, and for this audience it probably outperforms paid social by an order of magnitude.

Third — and this is the harder one — stop assuming organic search budget needs to grow proportional to traffic growth. For technical-buyer businesses, the rational move may be to hold organic Google investment flat and redirect the marginal pound toward earned media, technical PR, and the kind of authority-building work that gets you cited inside AI tools. The ROI math on that is genuinely hard to compute right now, but the ROI math on optimising harder for a shrinking channel isn't ambiguous. It's bad.

The bit nobody wants to say out loud

There's a version of this conversation that the SEO industry doesn't want to have publicly, which is: some audiences have moved, your client may be selling to one of them, and the honest answer to "how do we win in AI search?" is "be a brand worth citing, and stop expecting organic search to do work it can't do anymore."

That's not a satisfying deliverable. There's no dashboard for it. There's no monthly retainer line item that maps cleanly to "be more citation-worthy." It's also true.

The developer data is uncomfortable because it makes the audience question impossible to dodge. You can't look at a 299% Claude over-index and a 14% Google under-index and conclude that the right move is more content briefs. Something more fundamental has shifted, and pretending otherwise is what your competitors are doing.

The businesses that figure this out first won't have better SEO. They'll have a more honest read on where their buyers actually are.