Decision
Two linked decisions turn the open discoverability bet (G-055) from a hope into a posture, and settle whether OLN should block AI bots.
1. Citable, not trafficked — a destination, not an answer. Winning AEO means being the source answer engines cite; it does not by itself mean traffic. Answer engines are zero-click by design, so citation and visit move independently — you can win citation share and lose the traffic war in the same quarter. OLN therefore plays discoverability on two separate boards:
- Supply / citability — be the cleanest, freshest, Fact-grounded source an engine prefers to cite (single-domain authority, Entry 001; per-Fact provenance, G-026). This is the board OLN is unusually built for; Fandom's ad-choked, self-contradicting wikitext is its opposite.
- Demand / destination — convert the high-intent minority by owning the experiences a chat answer can't collapse into a paragraph: interactive timelines and relationship graphs rendered from graph edges, spoiler- and consumed-gated reading (Layer 10), canon-vs-speculation filtering, adaptive summary-vs-deep-dive — and participation itself. Nobody opens an answer engine to explore a universe. The reimagined world-model page (Entry 029) is that destination, and it is the only surface the traffic thesis can honestly stand on. The factual query the engine ate was never the click to fight for.
- Incumbency — the entrenched #1 and the zombie wiki are broken not by being marginally more citable but by rights-holder endorsement and canonical-source partnerships. That lever is promoted from a footnote to a load-bearing pillar (G-055).
2. Citable, not trainable. "AI bots" is not one switch; policy is per crawler class:
- Live citation / retrieval crawlers (OAI-SearchBot, PerplexityBot, AI-Overview fetch) — allow. They are the AEO traffic mechanism, and they attribute ("according to Loredoor…") — consistent with OLN's contributor-attribution principle.
- Training crawlers (GPTBot, CCBot, Google-Extended) — block by default. They bake contributor labor into model weights with no credit and no payment — the Fandom "monetizing volunteer labor without recognition" grievance (Entry 029) one layer up. Blocking them costs ~nothing in discoverability.
- Classic search (Googlebot) — allow. Table stakes.
The line OLN can say out loud: we let engines cite us, because citation credits our contributors; we don't let models train on that work for free, because that doesn't.
The licensable asset is the structured graph — Facts, normalized entities, relationships, provenance, freshness, the API — not the underlying CC-BY-SA prose, which cannot legally be fenced (G-044). So OLN stays freely citable now while holding the bulk-export / real-time / training tier in reserve as the thing a partnership actually buys.
Reasoning
The decisive correction is that AEO and traffic are decoupled. The earlier thesis quietly conflated "win the factual query" with "win the visit"; in a zero-click world those are opposite outcomes. Resting the funnel on un-answerable-in-chat experiences — explore, lurk, participate — is what survives answer engines eating the lookup.
"Block the bots and wait for a partnership once the data set is fuller" is half-right. The leverage half is real: training access, once given, can't be clawed back, and the structured graph is the crown jewel. But the waiting half has a chicken-and-egg flaw — you earn a partnership by being visibly the most-cited source, not by hiding; the density you'd wait for is itself built by discoverability → contributors. So the resolution is to seed citation now and hold the premium tier in reserve, not to go dark.
Splitting crawlers by class dissolves the apparent dilemma between "be the cited authority (G-055)" and "protect the asset (G-026, G-029)": citation and training are separable products sold to the same buyer. And the split is principled, not merely commercial — citation preserves attribution, training launders it — which makes it the commons stance (Layer 9, Entry 029) applied to crawlers.
The moat, restated once more: it is the graph, not the words. The prose is CC-BY-SA and free to anyone; the structure, provenance, and freshness are what OLN creates and can license.
Open threads
- Discoverability moat & migration (G-055) — re-sharpened around citable-vs-destination, split success metrics, and partnerships-as-pillar.
- AI crawler & data-licensing posture (G-058) — the operational policy: per-user-agent allow/block lists, WAF-level enforcement (robots.txt is honor-system), and the licensing-tier definitions. Depends on the CC-BY-SA derivative-scope opinion (G-044).