Getting-Found Launch Checklist: One Last Pass Before You Ship
Chapter finale. Everything from the first five lessons fits on one 12-item list. Each item is an action you can verify on the spot: inspect source, count entries, paste a URL, read a log. Check progress lives in this browser; come back every time you ship a new product. After the list, four measurement definitions answer the only question that matters: how do you know this work wasn't wasted.
Same as the Taste and Interaction lists: the right use is acceptance, not a memo. When a product is about to ship, walk the groups. Anything that fails is a todo. The four groups map to the chapter's four layers: crawlable and indexed, understandable, citable by AI, measurable. Most items are one-time jobs that stay done — that's why SEO is friendly to a one-person company.
Try it · Getting-found launch checklist: walk all 12
Getting-found launch checklist0 / 12
Crawlable & indexed · 4 items
Understandable · 2 items
Citable by AI · 4 items
Measurable · 2 items
This browser has local storage disabled (maybe private mode). Checks still work, but progress won't survive a refresh.
Honest note: this chapter's 6 pages were themselves accepted against this list, including "body visible on the server." A course about getting found that can't be found is the most embarrassing death this chapter can imagine.
Try it · How do you know it worked: four measurement definitions
Finishing the list only means "actions done." Effect lives on the meters. The four definitions are ordered by how fast they move. Open each card to see where to look, how often, and what healthy looks like. One more time, lesson 4's lesson: every number needs a date. A number with no definition is barely better than none.
Measurement meter cardsOpened 0 / 4
Ordered by how fast they move. Open them one by one.
AI crawler fetch volumeDaily
Where: your own request logs (after archiving by UA). Cadence: a glance every day. Healthy looks like: within a few days of a release or new content, GPTBot and PerplexityBot fetches step up. This is the most sensitive of the four meters — it's the first to know whether your content entered AI's field of view.
Index countWeekly
Where: coverage reports in Google Search Console and Bing Webmaster (more precise than site:). Cadence: log once a week, Google and Bing separately. Healthy looks like: index count climbs week by week toward your real page count, and the gap between the two engines is narrowing. If the gap suddenly widens, check whether sitemap or robots got broken.
AI-engine referral sessionsWeekly
Where: your own analytics, sessions whose referrer is a domain like chatgpt.com or perplexity.ai. Cadence: watch the share each week. Healthy looks like: from zero to some, then a slow rise in share. That's hard proof an AI citation brought a human — harder than any "we got cited" screenshot.
Target-question citation rateMonthly · North Star
Where: no ready-made report — run your own blind test. List 20 target questions; each month ask ChatGPT, Perplexity, Kimi, and Doubao (ByteDance's chatbot), and count how often your site is cited. Cadence: once a month, same questions, same engines. Healthy looks like: citation count rising month over month. Slowest, most annoying, and closest to GEO's real goal — that's why it's the North Star.
After all four, notice they form a causal chain: crawlers fetch (daily) → engines index (weekly) → AI brings sessions (weekly) → target questions get cited (monthly). When the upstream isn't moving, don't stare at the downstream and worry. Go back to the list.
Quiz · Which number deserves to be the North Star
The boss asks "how's our GEO going" — which number is the most honest answer?Single choice
All four are real numbers. Only one deserves to be the North Star.
ACumulative requests just passed ten million
BGPTBot fetched eight thousand times this week
COf 20 target questions, 7 AI answers cited us — last month it was 4
Dllms.txt is already live
Rhythm after launch: only four slots on the calendar
List checked, baseline logged — what's left is rhythm. Day-to-day maintenance for getting found is small enough to write as a fixed schedule. The point is to actually do every slot, and write the numbers into the tracker.
When
What to do
Output
Launch day
Walk the 12-item list, log a site: baseline, submit sitemap in webmaster tools
Tracker row 1: date plus each engine's index count
Daily, in passing
Glance at the AI crawler fetch curve
Anomalies (zero or a spike) caught the same day
Once a week
Log index counts (Google and Bing separately); check AI-engine referral share
One more tracker row, compared with last week
Once a month
Blind-test 20 target questions on four AI engines; pick one old article, add fresh data, republish
One North Star data point, plus one freshness signal
All of it is under half an hour a week. Getting found is not a project. It's a set of habits: actions go into scripts, numbers go into the tracker, topics go into the calendar. Keep all three on the rails and traffic is a function of time.
The tracker needs no special tool. Start a spreadsheet with five columns. Copy this:
Date
Google indexed
Bing indexed
AI crawler weekly fetches
AI referral sessions
2026-08-10
~693
~50
(start this week)
(start this week)
2026-08-17
…one row per week
Row 1 is this site's real baseline. Ugly, but written down. Put monthly blind-test results on a second sheet — 20 questions, one row each, check the ones that cited you.
Three lists, one pre-launch set
Here the third pre-launch list is complete: Taste covers looking good, Interaction covers working well, this one covers getting found. The first two inspect the product itself. This one inspects the product's connection to the world.
Myth-busting · Clear four obsessions before we close
Myth 1
"Do SEO after we have more content." Backwards. The infrastructure (the first ten list items) is cheapest when you have few pages. Once it's in the script, every new page ships already passing. Waiting until you have hundreds and then backfilling is the whole-site surgery from lesson 4.
Myth 2
"Indexed should mean traffic." Indexing is only layer two of the funnel. Indexed but no traffic usually means the show layer (nobody searches the question you answered) or the click layer (nobody wants to click the title). Go back to lesson 1's funnel. Don't keep reworking the infrastructure.
Myth 3
"GEO is a new discipline — I have to learn a whole second stack." You've already seen it in this chapter: ninety percent of GEO overlaps SEO. The extra ten percent (answer blocks, FAQ, llms.txt, dates) is all one-time work. Don't buy a course or a tool for it. It's on the list.
Myth 4
"The numbers look bad — don't write them down yet." Ugly baselines are exactly why you write them down. Bing at 50 pages in the tracker is what makes 300 pages two weeks later mean something. Optimization with no baseline is the same as not doing it. This site bought that discipline with one definition accident.
A launch prompt for AI
Same as the Taste and Interaction chapters: this list can be translated into a standing request for AI. When you start a new site or a big redesign, paste this:
Self-check before launch and report item by item: (1) key-page body text is visible in the HTML source; (2) each page has its own question-style title and description; (3) each page has a canonical; multilingual pages point at each other with hreflang; (4) generate sitemap.xml and robots.txt, no site-wide Disallow in robots; (5) inject FAQPage JSON-LD on Q&A pages; (6) under each H1, two or three self-contained summary sentences; (7) generate llms.txt at the root; (8) JSON-LD includes datePublished / dateModified. For each item, give a way to verify it.
Note the last line — "for each item, give a way to verify it." Make AI hand in homework with an acceptance path attached. You only spot-check. That's the right way to hand this chapter's method to AI.
Recap · Five lessons in three sentences
Diagnose
Lesson 1's funnel: no traffic? Split five layers and find the leak. Crawlable, indexed, shown, clicked — each layer has its own disease. A three-minute rough check (site: for index count, ask an AI once, search your brand) is always available.
Fix the base
Lessons 2–3, two battlefields: the SEO minimum-viable list gets engines to see you; GEO's four levers (answer blocks, FAQ, llms.txt, dates) get AI willing to cite you. All one-time jobs — they go into scripts, not the calendar.
Fight the long war
Lessons 4–5, real cases and content strategy: acceptance needs numbers and dates. The daily fight is topic selection: write questions people search, trade real experience for trust, stay away from black-hat.
The Bing-side door this chapter keeps stressing. Register, read the IndexNow integration docs while you're there, and you can finish the hookup in half a day.
A community proposal you can finish in ten minutes. The original sets much more accurate expectations than second-hand posts: what it is, what it isn't, who supports it.
Tools (the free tier is enough)
Google Rich Results Test / Schema Markup Validator
Two gates every structured-data write must pass. Search "rich results test" and "schema markup validator."
Your own request logs
Seriously: the best GEO tool most people ignore is raw access logs. Lesson 3's crawler roll call taught you how to read them.
The four AI engines themselves
ChatGPT, Perplexity, Kimi, Doubao — asking your target questions once a month is the most direct effect check. Perplexity's citation list also shows you who the competitors are.
Key takeaways
✓
Use the 12-item list as acceptance: crawlable and indexed, understandable, citable by AI, measurable — all four groups done means launched. Most items are one-time jobs. Put them in the build script and you're done.
✓
Read metrics as a causal chain: crawler fetches (daily) → index count (weekly) → AI referral sessions (weekly) → target-question citation rate (monthly North Star). If upstream isn't moving, don't stare at downstream and worry.
✓
Every number needs a date and a definition: "2026-08-10 measured Bing ~50 pages" is data. "We have lots of pages indexed" is vibes. The meters you keep for yourself follow the same rule.
✓
Three lists, one set: looks good, works well, gets found. The first two inspect the product itself. This one inspects the product's connection to the world.
✓
The acceptance request can go straight to AI: paste that launch prompt into the project rules. Make it hand in homework with a verification path. You only spot-check.
The chapter ends where lesson 1 started: Shipping is only the beginning. Getting found is what counts as launched. You now have a list, meters, and a tracker. Go put your product where it can be found.
Source: Original to Xiaoshan Academy's Getting Found series. List items all come from lessons 1–5 of this chapter. Measurement definitions and the blind-test method come from this site's actual GEO monitoring plan in August 2026.