Show up when the buyer asks ChatGPT. Substrate's AEO loop, ready to install.

This morning I measured my own brand against eight buyer prompts on AI search.
I appear once, on the vanity query, via third-party scrapers.
Substrate, my open-source GTM operating system, appears zero times.
Bluefish AI, the consulting shop I'd lose deals to in this space, sits in the slot I want.
The honest move is to publish the day-zero baseline before selling the engagement.

What it was

Search is splitting. The classic SERP that PMM spent fifteen years tuning against, and the AI answer that synthesizes three to seven citations into one paragraph the buyer reads instead of clicking through. Every SaaS founder I've spoken to since January asks the same question: when ChatGPT names the top three vendors in our category, are we one of them, and how would we know?

The diff

Before: SEO content programs built for crawl-rank-click, judged on impressions and organic sessions. AEO treated as a marketing-channel question, usually outsourced to whoever owned content.

After: AEO treated as a substrate question. Three primitives wired together: a relevance loop that tunes which claims the LLM will surface, a schema-delta read that tracks citation share against named rivals week over week, and a manual-action protocol for the case where the answer is wrong about your product.

What I actually did

1. Wrote skills/aeo-relevance in substrate. The relevance loop reads the buyer's category question, runs it through the three frontier models, parses citations, and scores who shows up against a named rival set. Output is a weekly delta, not a vanity ranking.
2. Wrote skills/aeo-tune in substrate. Translates the weekly relevance read into concrete on-page changes: claim ordering, schema patches (Organization, Product, FAQ, HowTo, Article), and source-page edits that move citation share against the rival set.
3. Wrote skills/aeo-manual-action. The protocol for the case the answer is wrong. Source-page update, schema patch, support-doc update, citation-trail to the corrected claim. Reusable across engagements.
4. Wired the three skills into one dispatcher so a Phase 1 engagement runs them as a single weekly routine. Output lands in the client's repo, not a vendor dashboard.
5. Set up the iamkesava.com baseline as the dry run. Established initial citation share for "pmm consultant india" and "ai-native pmm" prompts so the first client engagement compares apples to apples on the same instrument.

Day-zero baseline for iamkesava.com

I ran the loop on my own brand first. Eight prompts a buyer or founder would actually type into an AI assistant when shopping for consultants on AI-native PMM, GEO, or substrate work. One engine (Exa web search) as a proxy for generative-answer retrieval. Measured 2026-05-16. Raw results in data/aeo-baseline-2026-05-16.json.

Seven of eight buyer-shape prompts return zero appearances. The eighth, the vanity query, returns Kesava through RocketReach, Torre, Datanyze, and three LinkedIn posts. None of the owned domains (iamkesava.com, substrate.iamkesava.com, portfolio.iamkesava.com) appear anywhere in the result set. The category I claim to consult on is dominated by Bluefish AI, Matt McKinney, Ben Thiefels, Austin Heaton, Sam Awezec, and a long tail of fractional PMM and GEO agency sites with answer-first content shaped exactly for this read.

What stayed honest

One engine, web search as a proxy for the generative answer. A paid AEO product like Bluefish measures the full ChatGPT, Claude, Perplexity, Gemini, AI Overviews matrix with audience-segmented prompt panels. I don't, not yet. The baseline above was run on my own brand, not a client. Cadence from here: re-measure weekly on the same eight prompts and the same engine, watch the deltas. Phase 1 for a client is the same loop wired into their domain with the full multi-engine matrix and a named rival set. The HubSpot number at the top is still HubSpot's, not mine. Anyone selling AEO with retroactive metrics is selling you someone else's chart.

What it became, in substrate

• skills/aeo-relevance. Weekly citation-share read across frontier models against a named rival set.
• skills/aeo-tune. Turns the weekly read into on-page edits, schema patches, and claim reordering.
• skills/aeo-manual-action. Correction protocol when the LLM answer is wrong about your product.
• routines/aeo-routine. The engagement shape: measure first, install loop second, ship delta third. Weekly cadence, wired into one dispatcher.