In short: The first two passes are an afternoon of unglamorous checking, and they are where an audit should end if either one fails.
Reachability is answered by your logs, not by a tool's opinion: which AI crawlers actually reached you, how often, and what they got back. Extraction is answered by reading your own pages the way a system takes one passage out of them.
Both produce evidence somebody else can check, which is what separates these two passes from the two that follow.
Pass One: Reachability
Three questions, in this order, because each depends on the last.
Did they arrive
Filter thirty days of server logs for the user agents from module five: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended and the rest. Record which appeared, how often, and their status codes.
A crawler that never appears is either blocked, or has never found you, and those are different problems. Check robots.txt first because it is the cheaper explanation, then check whether anything links to the pages in question.
The finding that ends most audits early: a robots.txt written years ago, blocking a user agent nobody in the current team chose to block. It costs nothing to fix and everything to leave, and it will not appear in any tool that tests what a crawler could do rather than reading what one did.
Are the pages indexed
Google Search Console and Bing Webmaster Tools, both, for the same twenty pages. Bing is the one that gets skipped and it is the index behind Copilot, so a site well covered in Google and thin in Bing is invisible on a surface its team never checks.
Is the substance in the HTML
View source, not the rendered inspector. The inspector shows the page after JavaScript has run, which is the view a crawler may never get.
For each of the twenty pages, search the source for the sentence that answers the page's main question. If it is not there, the page's answer exists only after a script runs, and the fix is a rendering decision rather than a content one.
Pass Two: Extraction
This pass asks one question of each page: if a system lifted one passage out of it and had to defend using it, would that passage hold up alone?
The extraction check on this site will read a page back to you that way. It runs in your browser, uploads nothing, and produces findings rather than a score, because a score invites you to move the number instead of reading what it says.
What to do with the findings
Fix the top twenty pages and leave the rest. An extraction pass across four hundred pages is a project; across the twenty that carry your commercial questions it is a week, and those twenty are the ones a category question would retrieve.
Where This Pass Legitimately Ends The Audit
If the logs show crawlers arriving and the pages carry their answers in the source and near the top of their sections, both technical passes are closed and the audit moves on.
If either fails, write the finding and stop. The remaining two passes ask what an assistant says about you, and while your pages are unreachable or unreadable, the answer to that question is about your competitors rather than about you.
Record the negative findings too. "Bing coverage is complete" is worth writing down, because in three months somebody will ask whether it was checked, and an audit that only lists problems cannot answer that.
Key Takeaways
- Reachability is answered by logs, not by a simulation. A crawler that never appears is blocked or unlinked, and those are different problems.
- Check Bing as well as Google. It is the index behind Copilot and it is the one that gets skipped.
- Read the source, not the inspector. The inspector shows the page after JavaScript, which a crawler may never see.
- Extraction asks whether one passage holds up alone. The most common failure is a section that opens with setup instead of the answer.
- Fix the top twenty pages, not all of them, and record what passed as well as what failed.
Check yourself
Before you move on
Not scored, not recorded, and not part of the certificate. Both answers are settled by a sentence in this lesson, and the reasoning appears whichever option you pick.
- 01
Why do server logs beat a crawler simulation for the reachability pass?
- 02
Why check the page source rather than the browser inspector?