«We audited our own site before launching it.»
The cobbler's son problem applied к AEO consulting · self-AEO compliance was our hard launch gate.
ANSWER · AEO-EXTRACTABLE
We audited our own site before launching it. The cobbler's son problem in AEO consulting is real and unsolved across the category · we built vaeo.ai к not repeat it.
Most AEO vendors publish elaborate frameworks and have invisible-к-AI websites themselves. Their own sites lack llms.txt manifests · ship JavaScript-only content · run schema with 12 validation errors · have no sameAs network. They sell what they don’t apply.
Per DEC-WEB-005 P1 self-AEO compliance hard gate · vaeo.ai launch was blocked until we applied the full Citation Stack 4-layer audit к our own site · documented baseline measurements · executed every L1-L4 fix · ran cross-engine smoke test probe · preserved every artifact publicly.
Baseline (T0 · pre-launch · staging environment · 4 test queries · 5 engines): 0% Share of Model. Vaeo.ai didn’t exist yet · staging was crawler-disallowed · entity graph wasn’t established. Null state honest.
Seven first-pass misses disclosed honestly — three caught in initial self-audit during build · four caught in the comprehensive AEO compliance audit while writing the dogfood case itself. The fact that we caught four more while writing the case is the artifact: the audit catches what muscle memory misses · continuously.
Self-AEO probe (T-2 weeks pre-launch · 14 days post staging deployment): vaeo.ai cited in 3 of 5 engines (ChatGPT · Perplexity · Claude) on 3 of 4 test queries. Gate C threshold (≥2 of 5 · ≥2 of 4) met с margin. Aggregated Share of Model 35%. Gemini cited 0 of 4 (likely indexing window timing). AIO cited 0 of 4 (likely organic SERP not established). Partnership-geographic query cited 0 of 5 (likely L4 gap on entity-graph clarity).
The full audit artifact — baseline tables · Citation Stack composite scores · fix sequence atomic deliverables · crawler hit log · probe results · «what doesn’t work yet» disclosures · continuous measurement plan — lives at /cases/vaeo-applied-to-self.
Methodology proven against the methodology owner. Cobbler’s son has shoes.
N1 — The cobbler’s son problem
Most AEO vendors publish elaborate methodology and have invisible-к-AI websites themselves.
I’ve audited a lot of them. Not as paid client work — as background research before we built vaeo.ai. The pattern is consistent enough к be its own field note.
Their sites lack llms.txt. They ship JavaScript-only content (no SSR rendering · so AI crawlers see empty shells). They run schema with 8-12 validation errors and don’t catch it. Their sameAs network = single LinkedIn link. Their content sandwich pattern absent. They sell «Answer Capsule discipline» к clients while their own pages have no Answer Capsules.
They sell what they don’t apply.
The cobbler’s son problem is the pattern: the specialist with the methodology doesn’t apply it to themselves. In AEO consulting it’s particularly visible — the buyer can probe the vendor’s own site against AI engines and see the gap in 30 seconds.
When we started building vaeo.ai · we made a decision: self-AEO compliance is a hard launch gate. Per DEC-WEB-005 P1 (locked in the site spec) · the site doesn’t launch until we apply the full Citation Stack 4-layer audit к ourselves · document every fix · run cross-engine probe · preserve every artifact publicly.
The dogfood case is the artifact preventing the cobbler’s son pattern at vAEO. Methodology applied к the methodology owner.
«Most AEO vendors sell what they don't apply. The dogfood case is how we made sure we don't.»
N2 — Why we made self-audit a mandatory launch gate
The decision was DEC-WEB-005 P1 · written into the Pre-Launch Gate spec. Three criteria · all must pass before vaeo.ai goes live:
- AI crawler hit logs show 4 major AI bots visited key pages within 14 days of staging deployment
- Manual probe smoke test — vaeo.ai cited in ≥2 of 5 engines on ≥2 of 4 test queries
- Schema validation — Google Rich Results Test 0 errors across all 13 routes
No override. No «soft launch.» No «we’ll fix that after.» If all three don’t pass · launch is blocked.
Why hard gate · not soft target:
The methodology must apply к the methodology owner. If we can’t prove self-AEO compliance mechanically · we don’t ship. Anything less is the cobbler’s son problem applied к ourselves · which would invalidate every claim we make about other brands.
«If we can't prove self-AEO compliance · we don't ship. The hard gate is the discipline.»
N3 — The baseline · zero Share of Model
Pre-launch baseline probe · staging environment · 2026-05:
- 5 AI engines: ChatGPT (GPT-4o) · Perplexity · Gemini (Pro) · Google AI Overviews · Claude (3.5 Sonnet)
- 4 test queries: «What is verifiable AEO?» · «Who does AI Visibility audits?» · «Best Citation Stack methodology» · «AEO agency partnership Barcelona»
Result: 0/0/0/0/0 — vaeo.ai cited zero times in any engine on any query.
Aggregated Share of Model: 0%. P
Why the null state was the honest data point:
Vaeo.ai didn’t exist yet. Domain pointed к staging environment · crawler-disallowed by default · entity graph wasn’t established · no public sameAs network · no inbound links.
Zero Share of Model was the only honest baseline. Anything else would have been measurement theater.
What this looks like is what every newly-launched brand looks like in AI engines.
The dogfood case starts from this null state — which is where most of our client engagements also start. The starting condition is shared · the methodology applied is the same.
BBR started from 19% — established brand · already in AI engines’ training data · just badly. We started from 0% — brand didn’t exist yet. Different baseline · same methodology · same Citation Stack 4-layer audit.
«Zero Share of Model · five engines · four queries. The null state was the honest baseline. We didn't fake an early-mover advantage.»
N4 — Three first-pass misses (initial audit)
This is the trust-signal section. Most vendors publish dogfood cases framing everything as perfect from initial build. We name three first-pass misses honestly — because the methodology has к apply к imperfect execution · not idealized execution.
Miss 1 — llms.txt manifest absent.
Cot’s initial Astro 6 scaffold focused on robots.txt + sitemap.xml as the
canonical AI-discoverability layer. llms.txt wasn’t in the first build.
Caught in self-audit. Added к /llms.txt at root with canonical entity
overview · methodology summary · service ladder · contact pointer.
Time-to-fix: 2 hours.
Why we missed it: llms.txt is newer convention than robots.txt + sitemap.xml. Our own playbooks reference it · but initial scaffold-build attention defaulted к the older canonical set. The audit catches what muscle memory misses.
Miss 2 — Vlad’s Person schema missing GitHub URL in sameAs.
First draft Person schema included LinkedIn but not GitHub. Caught in second-pass schema audit. Added. Time-to-fix: 30 minutes.
Why we missed it: sameAs network was treated as «main social presence» in first pass · GitHub was treated as secondary. sameAs network = entity-graph completeness · not «main social profile.» Every authoritative cross-link counts at L4.
Miss 3 — Organization schema legalName + LocalBusiness streetAddress.
Most subtle miss. Early Organization schema included
legalName: "EMCC Digital SL" placeholder because the scaffold defaulted к
full LocalBusiness pattern with streetAddress. Both fields contradicted
DEC-WEB-001 canonical (no legalName · no streetAddress · only
addressLocality Barcelona). Caught in canonical review pass. Both removed.
Time-to-fix: 15 minutes.
Why we missed it: schema templates from Astro starter packs assume conventional LocalBusiness schema patterns. Our specific canonical decision (no entity declared per DEC-WEB-001) required manual override of default pattern. Canonical decisions require explicit checks against schema defaults · automation only catches what convention expects.
Plus four more caught while writing this:
The honest disclosure compounds. While writing the dogfood case page itself · a comprehensive AEO compliance audit caught four additional first-pass misses (UA-infrastructure-pre-mirror drift surfacing in three separate file locations · OG image broken link). Total: seven first-pass misses disclosed across the full dogfood case. The audit catches what muscle memory misses · continuously · including muscle memory formed by our own initial audit.
«Three first-pass misses · all caught in self-audit · all fixed before launch. Then four more caught while writing this. The audit is the discipline · not the execution.»
N5 — What still doesn’t work yet
The dogfood case isn’t a success story. It’s a snapshot of methodology applied honestly · including the parts that haven’t compounded yet.
Self-AEO probe results (T-2w pre-launch · 14 days post staging deployment):
| Engine | Cited |
|---|---|
| ChatGPT | 3 of 4 queries |
| Perplexity | 2 of 4 queries |
| Gemini | 0 of 4 queries |
| Google AI Overviews | 0 of 4 queries |
| Claude | 2 of 4 queries |
Gate C threshold met (≥2 of 5 engines · ≥2 of 4 queries). Aggregated Share of Model: 35%. G
Currently underperforming:
Gemini citation = 0 of 4. Gemini’s reliance on Google Search index means newly-launched sites face longer indexing window. Scheduled monitoring: weekly Gemini probe post-launch · expected citation appearance 30-60 days post-launch as Google indexing matures.
Google AI Overviews citation = 0 of 4. AIO uses ranked organic SERP positions as content base. vaeo.ai SERP rankings hadn’t established at pre-launch probe. Scheduled monitoring: monthly AIO probe · expected citation 60-90 days post-launch.
«AEO agency partnership Barcelona» cited 0 of 5. L4 gap — entity-graph hadn’t established «Barcelona» + «partnership» + «AEO» cross-reinforcement at pre-launch. Scheduled L4 fix: Wikidata entity creation · industry directory placements · Substack publication launch · LinkedIn company page activation.
What success looks like in 90 days (post-launch target):
- Aggregated Share of Model ≥ 50% across 5 engines on 4 test queries (vs 35% at launch)
- Gemini citation ≥ 1 of 4 queries
- AIO citation ≥ 1 of 4 queries
- L4 composite Citation Stack score ≥ 85 (vs 71 at launch)
The continuous measurement plan:
Same monthly cadence vAEO sells к clients · applied к ourselves. Monthly cross-engine probe · drift detection · proactive fix work on drift · monthly written report. Public updates к dogfood case page: Q3 2026 + Q4 2026 + annual retrospective. Honest about both compounding and failures.
The measurement framework follows Princeton GEO research (Aggarwal et al. 2024) on AI engine citation rate optimization — cross-engine probing · Statistics Addition · Quotation Addition · Cite Sources discipline applied to our own pages.
«35% at launch. 50% target at 90 days. Gemini + AIO + partnership-geographic still work к do. We name what doesn't work yet · we don't fake the chart.»
N6 — Where this leaves us · and you
The dogfood case is the load-bearing answer к cobbler’s son problem in AEO consulting. Methodology applied к the methodology owner. Baseline preserved. Audit artifact public. Fix sequence documented atomically. Probe results honest. Limitations named.
For us: continuous monthly cadence post-launch. Public updates · including failure modes. The dogfood case page becomes a live measurement instrument · not a launch artifact frozen в time.
For you: if vAEO methodology can produce Citation Stack composite 86 (Gold-grade) on a launching site · с full audit transparency · including seven first-pass misses disclosed publicly · the same methodology applied к your brand is the work you can audit. Same SKU. Same Citation Stack 4-layer model. Same Proof Ladder framing.
Where to go next:
«Cobbler's son has shoes. Methodology proven against the methodology owner. If we don't cite ourselves when AI is asked about AEO · we have no business charging for it.»