DEC-0073 — Visibility & GTM: fix the floor before widening the mouth

Reversibility: two-way door — this picks a sequence and a posture, not an irreversible mechanism. Every "YES-LATER" (blog, social, llms.txt) can be started at any time; the deliberate choice is to not start them yet. The durable commitment is the principle (sequence by the funnel; comprehension before awareness) and the anti-goals — both revisable when the floor is proven to hold.

Founder-directed this session, steered through a four-lens workflow (Product Owner, Principal Platform Architect, Documentation Engineer & Information Architect, a growth-strategist lens). Executes on the register topology locked in DEC-0072. confidence: asserted — a forward GTM bet grounded in a verified in-repo audit and live 2026 growth research; the conversion outcome is unproven. Recorded automatically in the same turn (the standing rule).

Context

The founder asked, in effect, "how do we get seen?" — surfacing five concrete questions (official product docs; SEO/AEO/GEO; llms.txt; GTM motion; social/blog) and one explicit worry (does AI-assisted content get penalized?). Two grounding inputs frame the answer:

• A verified in-repo audit of our actual surfaces. OG/Twitter cards are absent on every surface; canonical is absent on all SvelteKit routes; robots.txt, sitemap, llms.txt, and JSON-LD are all absent; and the only docs surface today renders our own ADR log — a stranger who lands there bounces. The proof that a builder-ICP converts on (a runnable offline loop + a strict-validated Claude Code plugin) exists in code but is undocumented as a product.
• Live 2026 growth research. Off-site / third-party signals (brand mentions, G2/Capterra, Reddit/Wikipedia) are the strongest predictor of AI-engine citation and compound over 30–90 days; llms.txt adoption is ~10% with ~no measured citation lift; and search/answer engines do not penalize AI-assisted content as a category — Google's 2026 scaled-content-abuse updates punished thin programmatic page-farms (50–80% traffic drops), i.e. scale without value, not AI involvement.

The binding fact: we have no conversion data yet (DEC-0022's review flags product-market fit open). So awareness spend is a blind bet until the surface itself converts.

The principle (the new direction)

The binding constraint is COMPREHENSION + ACTIVATION at the surface, NOT awareness. Sequence strictly by the funnel: fix the floor before widening the mouth. Build the on-ramp (official product docs) + cheap shareability/crawlability hygiene first; every awareness instrument (blog, social, SEO content, llms.txt, off-site seeding) is downstream and is wasteful-or-dishonest before the asset it amplifies exists. Driving traffic to a surface that doesn't yet explain or activate the product burns the most expensive thing a near-zero-distribution team has — the first impression — against an unfinished floor.

Decision — verdicts on the founder's five questions

1. Official product docs — YES-NOW (P0, the keystone). The single highest-leverage gap: the conversion-bearing asset (the runnable loop + the plugin) is undocumented. Built as the separate authored /docs/* register locked in DEC-0072. Filed: Stand up the product-docs register — REPURPOSE the @dossier/docs Starlight surface to official product docs at /docs/* (new-tab "Docs").
2. SEO / AEO / GEO — SPLIT.
   • Hygiene = YES-NOW / P0. OG + Twitter cards, canonical on the SvelteKit routes, a sitemap, robots.txt with an AI-crawler allow-list, and Organization + SoftwareApplication JSON-LD. This is table-stakes crawlability/shareability, built once. Filed: Shareability + crawlability hygiene pass — one atomic <SiteHead> (OG/Twitter + canonical + JSON-LD), robots.txt, sitemap.
   • A keyword/content SEO program + a managed GEO play = NO. Premature for a tiny team with no conversion floor and no off-site authority yet; a content/keyword engine is an awareness instrument (downstream), and a managed GEO program is spend against an unproven floor.
3. llms.txt — YES-LATER. Honestly low-ROI (~10% adoption, no measured citation lift) but on-thesis: a Business-to-Agent / paste-into-Claude channel, derived from the OKF/docs SSOT, never hand-maintained and never sold as a ranking lever. Ships after docs exist (it derives from them).
4. GTM — YES-NOW, but a product-led + agency-CHANNEL proof motion. Per DEC-0023 (agencies are the channel, not the gate) and DEC-0052 (dogfood-first, land-and-expand). Concretely: finish the DEC-0014 plugin-marketplace distribution (the highest-leverage underused channel — the plugin is built but not verified as an installable product) + hand-recruit a small design-partner cohort + harvest each loop as a case-study artifact atom. Explicitly NOT: a pricing page, paid ads, or a sales team.
5. Social — YES-LATER. Footer/contact links + OG now (part of the hygiene pass); a narrow, founder-led build-in-public on X later, drawing only on already-public honest assets — the decision log, the live /board, the /activity ledger (The WATCH face — a read-only compounding-return ledger over the board snapshot + the tenant git trail). Not a multi-platform calendar.
6. Blog — YES-LATER. 4–6 canonical, depth/citable essays (the /blog/* register of DEC-0072), NOT a cadence treadmill. Depth over volume; each essay is an asset, not a content-mill unit.

Distribution posture (recorded so the generic SEO/marketing checklist stops re-litigating it)

• Hygiene metadata is built ONCE as atomic shared head components, fed from existing SSOTs — the landing brand.meta* model for the SvelteKit app, Starlight frontmatter for docs. No per-page hand-authored meta; one <SiteHead> (shared-chrome discipline).
• llms.txt / sitemap / feeds are DERIVED views over the SSOT, never hand-maintained. They are generated from the atoms/frontmatter, so they can't drift.
• Analytics stays privacy-friendly, Vercel-only (DEC-0027) — no third-party trackers.

AI-generated-content guardrail (the founder raised this explicitly — captured here)

Search/answer engines do NOT penalize AI-assisted content as a category; they demote SCALED low-value content (Google's 2026 scaled-content-abuse updates hit thin programmatic page-farms with 50–80% drops). The line is scale without value, not AI involvement. Therefore, non-negotiably:

• Human-curation gate before publish — the agent drafts, a human disposes (the same review-gate pattern as the loop, Governed disposition lives with the operator control plane (CLI now, local dashboard optional, deployed siloed service deferred) — the public `/board` deploy stays read-only and holds no mutation authority).
• E-E-A-T by construction — a real founder byline; the dogfood KB as genuine evidence of experience (we run the loop we sell); provenance where it fits.
• Depth over volume, and CURATE-DON'T-SCALE: never turn the dogfood KB (or any client KB) into a programmatic SEO page-farm. The KB is safe ONLY as human-curated, genuinely-valuable-per-atom content — the exact opposite of the page-farm pattern the algorithms punish.

Explicit anti-goals at this stage

Premature for a near-zero-distribution tiny team, and several contradict the sovereignty/privacy posture:

• a pricing page;
• paid acquisition;
• a multi-platform social calendar;
• a content/keyword SEO engine;
• a managed GEO program;
• third-party trackers (GA / Segment / Mixpanel — contra DEC-0027);
• any programmatic page-farm over the KB (contra the guardrail above and DEC-0072's KB-purity invariant).

Rationale

• The audit makes the sequence obvious, not arbitrary. Hygiene is absent and the conversion asset is undocumented — so the cheapest, highest-leverage moves are the floor (docs + hygiene), and any awareness instrument pointed at the current surface would amplify a 404-of-comprehension. Fix the floor first.
• The research says awareness is the slow, compounding, off-site game — not the near-term lever. Off-site authority compounds over 30–90 days and is what AI engines actually cite; llms.txt shows no measured lift. So spending now on awareness instruments is both premature (no floor) and low-yield (the durable signal is off-site reputation we earn by being genuinely good, over months).
• On-thesis distribution only. Every YES here is owned and derived — docs and hygiene fed from our own SSOTs, llms.txt/sitemap derived, social/build-in-public drawn from already-public honest surfaces, GTM via the plugin channel and real design-partner loops. Nothing here is rented attention or fabricated authority; the moat is the loop itself made legible (One loop, three faces — ship an agentic Digital Experience Agency "in a box" (the Agency) plus an adoption-ladder readout product (the Ladder) over the OKF Company Brain, dogfood-first, Dossier — Mission & North Star).
• The guardrail protects the asset. Our single most credible growth asset is the dogfood KB as genuine evidence. Scaling it into an SEO page-farm would both trip the 2026 scaled-content-abuse line and destroy the very credibility (DEC-0072's purity) that makes it convert. Curate, never scale.

Consequences

• Two P0 "Now" moves are filed as board tasks so the work doesn't evaporate (Agentic "sprint board" architecture — a git-resident OKF task board worked by bounded, hook-governed Agent SDK loops, Dossier — The Knowledge Model (v0)): Stand up the product-docs register — REPURPOSE the @dossier/docs Starlight surface to official product docs at /docs/* (new-tab "Docs") (the keystone, depends on DEC-0072) and Shareability + crawlability hygiene pass — one atomic <SiteHead> (OG/Twitter + canonical + JSON-LD), robots.txt, sitemap. The "Next" comprehension-funnel CTA + the DEC-0014 plugin-install verification are filed too: Comprehension funnel — route the landing/docs surfaces into a clear primary CTA toward the on-ramp, Finish DEC-0014 plugin distribution — verify the Claude Code plugin installs as a real product from the marketplace.
• The generic "do all the SEO/marketing things" checklist is now settled, not re-opened. The split verdict and the anti-goals are the standing answer; a future "should we buy ads / build a keyword engine / run a managed GEO program?" is answered "no, until the floor is proven" — revisit when conversion data exists.
• Sequencing dependency recorded: docs (P0) precedes llms.txt, blog, and the awareness instruments — they derive from or amplify the docs/KB SSOT, so they cannot meaningfully precede it.
• No new analytics or trackers — the demand signal stays the DEC-0027 privacy-friendly Vercel-only path plus the owned Buttondown list (Ship the landing publicly behind a docs-gate flag, capture demand through two honest doors, into a list we own).

Review

status: active (founder-directed; the posture is in force) · confidence: asserted (a forward GTM bet — grounded in a verified in-repo metadata audit and live 2026 research, but the conversion outcome is unproven; there is no PMF data yet per DEC-0022). Promotion gate (asserted → verified): the floor is built (the Stand up the product-docs register — REPURPOSE the @dossier/docs Starlight surface to official product docs at /docs/* (new-tab "Docs") + Shareability + crawlability hygiene pass — one atomic <SiteHead> (OG/Twitter + canonical + JSON-LD), robots.txt, sitemap acceptance criteria met) and the privacy-friendly funnel (DEC-0027 + Buttondown) shows a measurable lift in comprehension/activation (docs reach + early-access conversion) attributable to the floor — at which point the YES-LATER awareness instruments graduate from "premature" to "sequenced next" on evidence, not assertion. Until then this records the sequence and posture, not a proven outcome.