Supermetric System: Testing AEO Metrics That Actually Work

Written by John Steele | Jul 8, 2026 10:08:34 PM

For twenty years, an SEO tool did one genuinely useful thing: it told you where you stood. Point Moz or Semrush at your domain and it ran you against a bounded universe of keywords with real, measured search volume, then handed back every term you ranked for -- including ones you'd never have thought to check. It even rolled the whole picture into one number, Domain Authority, a rough but mildly deterministic score that anchored SEO reports for a decade. None of it was perfect, but it gave you something to test against.

AEO tools promise the same move for ChatGPT and Perplexity. But here's the truth: they structurally can't deliver it.

There is no bounded universe of prompts. They're infinite, conversational, different every time -- so instead of showing you where you stand, the tools ask you to guess which prompts to track, then bill you per guess. And the guesses don't hold. Profound found ChatGPT rewrites 91% of prompts into different internal queries before it retrieves anything, so the prompt you're paying to monitor isn't the one the model runs. A single chat is a chain of ten such prompts, none of them tracked.

In January, Rand Fishkin--the guy who invented Domain Authority--ran the same recommendation prompt through the LLMs a hundred times and got the same list of brands less than once.

So prompt tracking can't tell you where you stand, because there's no stable "where" to measure. That's a problem, because the money is sprinting toward exactly that -- Adobe paid $1.9 billion for Semrush, HubSpot paid $30m for xFunnel to ship a $50-a-month AEO product, every CMO survey has budgets moving to answer engines this year.

So I tried to build the thing AEO is missing: a rough, benchmarkable supermetric -- a Domain Authority for the AI era. Here's how.

	Our W2W is a single 0–100 read on the three things AI actually weighs: third-party mentions, branded search, and reviews. Octolens, Google Search Console, and your category's review sites, piped
	through Make or Zapier into one Google Sheet. About $159 + an afternoon to stand up.

Years ago, Rand Fishkin and Moz invented a supermetric that anchored SEO reports for a decade. Now Fishkin's new company has published the research showing we need a replacement.

Underneath the random rankings, Fishkin found the consideration set is stable: ask about headphones a hundred times and Sony, Bose, and Apple appear in 55 to 77% of answers regardless of phrasing. The rank is noise; membership is signal. And the research is clear on what earns membership.

Branded Search:

Ahrefs found the majority of ChatGPT's most-cited pages come from high-authority domains.

Mentions:

Muck Rack's study of over a million prompts found most non-paid AI citations come from earned media, not your own site -- and its data shows brand mentions outpredict backlinks for AI visibility.

Reviews:

A Trustpilot/Seer analysis of 800,000 answers found brands with no reviews were cited around 1% of the time; brands with 80-plus managed reviews, roughly 75%.

AI recommends the brands the rest of the internet already talks about, searches for, and vouches for. So that's what we measure.

› One number, benchmarked against your tracked competitors, built from data you can pull instead of a probabilistic lottery. "We're a 55, the category leader's a 99" is a sentence you can act on.
› A way to directly measure what gets brands into the consideration set -- mentions, branded search, reviews -- not the rank that reshuffles every time someone hits enter.
› A dynamic report of top metrics you can pipe into any dashboard.

› A real-time read on new reviews--a metric you can directly influence.

Prerequisites:

Make or Zapier

Middleware makes this a very straightforward, no-code build. All tools integrate with them directly. Just connect all your data sources and use the middleware to pass the data to Google Sheets.

› Pull mentions: Octolens offers API access is on every plan (no enterprise upsell). It is purpose-built to track AI visibility-- Reddit, Hacker News, GitHub, LinkedIn, YouTube, podcasts -- and one POST to its /api/v2/mentions endpoint returns clean JSON where each mention already carries what the score needs: an AI relevance score to filter noise, plus engagement metrics and follower count to do the reach-weighting. A mention in a 50k-view thread outweighs one in a dead post. Filter by relevance first, then weight by engagement.

› Get branded search: Us the Google Search Connector in your middleware to send your branded-query volume -- name, products, key people -- with a category-term row for context. This is the Share-of-Search input.

› Decide on the best review site for your brand category and connect: From your category's platforms -- G2 and Capterra for B2B software, Trustpilot or Google for service businesses -- pull count, average rating, and month-over-month change.

› Normalize each input: Create and connect your Google Sheet. On a Score tab, score every input 0–100 against whoever leads it. If column B holds each brand's reach-weighted mention total: =ROUND(B2/MAX(B:B)*100, 1). Same for search. The leader lands at 100; everyone else is a share of them.

› Blend the reviews input: Nobody should win on three five-star reviews -- volume 70%, rating 30%. Count in B, average rating (out of 5) in C: =ROUND(0.7*(B2/MAX(B:B)*100) + 0.3*(C2/5*100), 1).

› Weight-sum the score: Weights in a header row -- 0.45 mentions, 0.33 search, 0.22 reviews -- then: =ROUND(SUMPRODUCT(B2:D2, $B$1:$D$1), 1). Chart that column over time and you've got the whole instrument.

› Quarterly prompt check: Check some prompts across the platforms, making note when a competitor is mentioned. Use these prompts to test the score as you move forward.

It won't tell you you're winning AI search -- nothing honestly can yet. But it will tell you whether the things that drive AI recommendations are trending your way, months before revenue would show it. Run it a few quarters against your pipeline. If it holds, you've got your metric. If it doesn't, you learned that for the price of an afternoon -- which is more than people paying $10k+ into AEO this year can say.

Full disclosure: Projects like this are what JSC is here for. If you want a hand standing it up, book a 30-minute intro call.

Salesforce agreed to acquire Fin -- the AI customer-service platform formerly known as Intercom -- for approximately $3.6 billion on June 15, its largest deal since Slack. Fin's agents resolve support queries across chat, email, WhatsApp, and Slack, and will fold into Agentforce. The deal is expected to close in Salesforce's fiscal year. Learn More...

Documentation platform Mintlify acquired Helicone, the open-source LLM observability tool, on March 3. Helicone had processed 14.2 trillion tokens across 16,000 organizations. The sale of an unmitigated success story in the space has some wondering about the future of AI observability as a standalone product area. Learn More...

HubSpot introduced HubSpot AEO at its Spring 2026 Spotlight, giving marketers a dashboard for how often a brand appears across ChatGPT, Gemini, and Perplexity. It's built on the company's ~$30 million October acquisition of xFunnel, and lands as HubSpot pushes its "Loop Marketing" playbook around AI-driven discovery. Learn More...

View full post