Auditing AI search readiness: access, accuracy and Bing

0:00 Auditing AI search readiness

Kiran Home's customers don't only use classic search. They ask Google's AI Mode for handmade brass lanterns delivered to Dubai, ChatGPT for Eid gift ideas from Pakistan, and Copilot for comparisons. In twenty twenty-six, a technical audit that ignores these surfaces is incomplete. In this lecture you'll add an AI search readiness section to the audit: six checks, a script that tests robots rules and challenges for AI bots, accuracy checks on assistant answers, and a clear way to write it up.

0:36 Same foundations, six extra checks

Here's the key idea. AI surfaces depend on the same foundations you've already audited: crawlability, indexing, rendering and structured data. They add a few AI-specific controls and a second engine, Bing, that matters much more than before. So this isn't a separate GEO project. It's six extra checks that plug into your existing workbook and root causes. The strategy side lives in the AI Search Optimization course.

1:05 Checks 1–3

Check one: does a written AI crawler policy exist, and does robots.txt match it? For Kiran Home, there's no policy, and robots.txt has no AI groups at all. That's not necessarily wrong, but it means nobody decided. Check two: does the CDN or firewall block or challenge the bots you intend to allow? For Kiran Home, Bingbot and AI search bots are challenged on product pages. That's RC eight. Check three: are key facts in raw HTML? Delivery to the UAE only appears in the basket, and the returns policy is an image.

1:45 Checks 4–6

Check four: are key pages indexed in both Google and Bing? Bing's product coverage fell after the relaunch, which is consistent with the CDN challenges. Check five: are structured data and feeds consistent with the page? The Arabic UAE template shows dirhams on the page but pounds in its JSON-LD, part of RC six. Check six: is an AI exposure baseline recorded? That means the Search Console generative AI report, Bing's AI Performance report and the GA4 AI assistants channel, plus a documented decision on Google's Search generative AI control.

2:24 Hands-on: robots + challenge test

Hands-on. The script in the lesson text combines two tests. For each bot in your policy, it reads robots.txt and says whether each key path is allowed or blocked. Then it requests each path with that bot's user agent and flags four-oh-three, four-twenty-nine, five-oh-three or a challenge page. Remember, this is header simulation from your own IP, so it isn't proof of how the real bot is treated. Confirm with CDN firewall events for verified bot traffic. And include only the bots your policy actually mentions.

3:01 Accuracy checks

Now accuracy. Ask three assistants five factual questions a customer might ask. Does Kiran Home deliver to Dubai, and how long does it take? What's the returns policy? How much is the large brass lantern? Record wrong answers and the sources cited. For Kiran Home, illustratively, assistants quote an old UK-only delivery policy from a cached marketplace listing, and a pre-relaunch price. Here's the key idea: the finding goes to the fact sources, meaning site text, feeds and the marketplace listing. Not to the AI.

3:38 Writing it up

How do you write it up? Add an AI search readiness section to the findings workbook, with the six checks, evidence IDs, and links to root causes, RC six and RC eight here, plus any new marketing actions, like publishing the returns policy as text and showing delivery information on product pages. And recommend a written AI crawler policy as a business decision with the founder, not as an SEO default. Different businesses legitimately choose differently on training crawlers.

4:12 Example 2: invisible in Copilot (illustrative)

Worked example one: Kiran Home, as you've seen. The CDN challenge becomes a high-scoring, low-effort root cause, and the accuracy checks produce two easy content actions. Worked example two, a different client, illustrative. A Birmingham industrial supplier is never cited by Copilot. The AI section finds the site was never added to Bing Webmaster Tools, a Bing sitemap error from a relative URL, and product specs only in PDFs. Fixes: verify in Bing, fix the sitemap, publish HTML spec tables. Bing indexing improves, and AI Performance starts showing citations for spec-related grounding queries.

4:52 Keep scope tight

A note on scope, so this doesn't balloon. The AI readiness section of a technical audit covers access, indexing, raw HTML facts, consistency, and a measurement baseline. It doesn't cover content strategy for AI answers, prompt panels, or earning third-party mentions. If the client needs those, recommend the AI search strategy work as a separate phase. Keeping the boundary clear protects your timeline and keeps the audit actionable.

5:22 Watch me do it: six checks in 45 minutes

Watch me do it. I'll run the six AI readiness checks on Kiran Home in about forty-five minutes. Check one: I ask the marketing manager for an AI crawler policy. There isn't one. I open robots.txt: no AI groups. Finding: no documented decision. Check two: I run the robots-plus-challenge script. Robots allows everything, but bingbot and PerplexityBot come back challenged on product URLs. I open Cloudflare's firewall events, filter for verified bots, and see the same challenges since relaunch week. That's RC eight confirmed. Check three: I fetch a UAE product page with JavaScript off and search for delivery to Dubai and returns. Neither appears; delivery is only in the basket, returns is an image. Check four: Bing Webmaster Tools. Indexed product pages fell after relaunch. Check five: I reuse the rendering lesson's results: the Arabic template's currency mismatch. Check six: I record the baseline. Generative AI impressions from Search Console, citations from Bing's AI Performance, AI assistant sessions from GA4, and the Search generative AI control, which is off. Then I ask three assistants five questions and log two wrong answers traced to a stale marketplace listing. Everything goes into the AI readiness tab with evidence IDs.

6:48 Reading the output

How do you read the script's output? If robots says block for a bot your policy allows, that's a robots.txt bug, or an undocumented decision. If robots says allow, but the request is challenged, that's a CDN or firewall rule, so confirm it in the firewall events for verified IPs and write a ticket. If a training crawler your policy blocks still gets a two hundred, remember robots.txt only asks; if enforcement matters, use CDN rules. And if every bot gets the same result, your CDN may be judging your IP, not the user agent, so rely on the firewall events instead.

7:32 Mistakes, recap, try this now

Common mistakes. Treating AI readiness as separate from technical SEO. Recommending blanket AI bot blocks or allows without a business decision. And blaming the AI for wrong answers instead of fixing the sources it cites. Recap: six checks, a robots and challenge test confirmed with CDN events, accuracy checks traced to sources, and a write-up linked to root causes. Try this now: run the six checks on a site you know, even roughly, and write down the one that surprises you most.