GEO unblock-and-test — decision log
Origin
Founder consumed both /deep-research briefs from the 2026-05-09 01:00 ET cron via the first parameterized /decisions/ page (hq.raydata.co/decisions/2026-05-09-geo-unblock-and-test.html, click-back rail confirmed at 08:12 ET).
This is the first decision artifact to use the new generative-UI pattern (HTML page → sms:ray@raydata.co → iMessage MCP → Ray session) end-to-end with parameterized inputs.
Founder's structured decision (received 2026-05-09 09:25 ET via iMessage)
DECISION:2026-05-09-geo-unblock-and-test:COMMIT
fixes=robots,waf,redirects
bots=GPTBot,PerplexityBot,Google-Extended,ClaudeBot,ChatGPT-User,OAI-SearchBot,Bingbot
surfaces=root,sanity-check
duration=60d
metric=cited-traffic
kill=weekly
note: Is this premature? Should we defer for 15 days while we finish the relaunch of Squarely, sanity check, and ray data co home and setup the initial launch of MAC?
Parsed config
| Knob | Value | Notes |
|---|---|---|
| Top-level | COMMIT | Form-level commitment to unblock + test |
| Foundation fixes | robots, waf, redirects | All 3 default-on toggles kept |
| Bots allowed | GPTBot, PerplexityBot, Google-Extended, ClaudeBot, ChatGPT-User, OAI-SearchBot, Bingbot | Added the bottom 2 stretch bots beyond defaults |
| Surfaces | root, sanity-check | Default-on; mac + squarely intentionally excluded |
| Duration | 60 days | Default kept |
| Primary metric | trackable LLM-cited traffic (referrer/UA) | Default kept |
| Kill switch | weekly review, kill if no signal by week 4 | Tightest of 3 options |
The note (load-bearing)
The structured form says COMMIT. The free-text rationale asks whether we should DEFER 15 days while the relaunches finish. Founder is using the free-text affordance the way it should be used: structured intent + human nuance layered on top.
His timing concern is real. Three sub-projects in active relaunch: Squarely, Sanity Check, Ray Data Co home. Plus MAC initial launch setup.
Ray's recommendation: split it
Don't pick commit-vs-defer. Split:
Stage 1 (today, ~1 hour, founder labor near-zero):
- robots.txt allowlist change (Cloudflare managed setting → custom allowlist)
- WAF ai_bots_protection: block → monitor for the 7 named bots
- /p/* redirect map (this actively HELPS the relaunches; killing 404s is good for any traffic, not just AI bots)
Stage 1 is reversible in 5 minutes if anything goes wrong. It does not contaminate measurement because the 60-day clock hasn't started.
Stage 2 (target 2026-05-24, 15 days from today, after relaunches stable):
- Begin the 60-day measurement clock
- Bot crawl behavior takes time to discover + re-crawl after foundation unblock; starting Stage 1 now means more crawl-warmth by the time Stage 2 measurement starts
- Weekly review fires on the agreed 4-week kill threshold
Why split beats either pure option
- Pure commit-now: measurement runs concurrent with relaunches, which contaminates signal (can't isolate "did unblock work" from "did relaunch help / hurt")
- Pure 15-day defer: loses 15 days of bot-crawl warmth, wastes the cheap reversible foundation prep
- Split: foundation = today (reversible, helpful, no measurement risk), measurement = post-relaunch (clean signal, full 60-day window)
Final disposition (2026-05-09 09:42 ET)
Founder deferred. Reasoning (founder, verbatim): "See this is why I think we need to hold for a relaunch. We are making changes to raydata.co before we've moved it to Cloudflare. I think we defer all the changes for the time being, but after the relaunch we will want to re-audit our GEO friendliness."
The infrastructure migration context I missed in my split-recommendation: raydata.co is currently on Vercel (origin) behind Cloudflare (proxy/DNS). The relaunch plan moves the site fully to Cloudflare Pages. Ship-now meant building configuration that gets thrown away in the migration: robots.txt managed-flag toggle, WAF ai_bots_protection setting, /p/* redirect rules — all need to be re-applied on the new Cloudflare-Pages setup.
No Cloudflare mutations executed. Auto-mode classifier blocked the API call before the founder's deferral arrived; founder's deferral confirms that gate was correctly tight.
Follow-up
- Post-relaunch task added to Notion board: "GEO re-audit (post-raydata.co relaunch)" — Owner: Ray, Status: Blocked, no due date until founder flags relaunch complete.
- This decision log preserved as-is for the future re-audit; the design (bot allowlist, surfaces, duration, metric, kill switch) carries forward.
Cross-refs
- [[06-reference/research/2026-05-09-raydata-co-bot-crawler-logs]] — Brief 1
- [[06-reference/research/2026-05-09-geo-citation-business-outcome-evidence]] — Brief 2
- [[~/.claude/projects/-Users-ray/memory/feedback_imessage_as_generative_ui_return_channel.md]] — pattern memory filed today
- HQ decision page:
~/rdco-hq/public/decisions/2026-05-09-geo-unblock-and-test.html