The AI search checklist, in priority order.

Checklists in this discipline usually list everything as equally important, which is unhelpful because the items differ in return by an order of magnitude. This one is ordered.

Tier 1: do these first

• Baseline measurement. Twenty buyer questions, three assistants, results recorded. Everything else is guesswork without it.
• Entity consistency audit. Legal name, trading name, address, phone and services identical everywhere they appear.
• Organization schema, complete. Including sameAs links to profiles that corroborate you.
• Crawler access check. Server logs and robots.txt. Unintentional blocking is more common than people expect.
• Source list. Which domains supply answers in your category.

How to run the tier 1 baseline in an afternoon

Write twenty questions a buyer would ask before choosing a business like yours. Mix three kinds: recommendation questions (“best roofing company in Tacoma”), comparison questions (“metal vs asphalt roof cost”) and problem questions (“roof leaking after a storm, what should I do”). Use the words customers actually use on calls.

Ask each question in ChatGPT, Google (looking at the AI Overview or AI Mode answer) and one of Gemini, Perplexity or Copilot. For each answer, record whether you are named, who else is named, which URLs are cited, and anything inaccurate. Use a logged-out or fresh session where possible, and note the date. That sheet is your baseline, and every later decision should be checked against it.

Tier 2: high return, moderate effort

• Restructure your top commercial pages for extraction: question headings, answers first, self-contained sections.
• Service and FAQ schema on the pages that matter, validated as JSON.
• Correct your listings on the sources from your tier 1 source list.
• Author identity made consistent across your site and professional profiles.
• Fix stale information anywhere it appears, including on sites you had forgotten about.

Tier 3: worth doing once tiers 1 and 2 are done

• New content aimed at measured gaps rather than at keyword lists.
• Outreach for representation on sources you are absent from.
• Ongoing monitoring cadence with a fixed prompt set.
• Comparison and alternative content, which assistants use heavily.
• llms.txt, if you have an hour spare and nothing in tiers 1 or 2 outstanding.

What is not on this list

Publishing at volume, keyword density, and anything promising to influence model training. The first dilutes quality signals, the second has not mattered for a decade, and the third is not achievable by anyone selling it to you.

Tier 1 technical checks, specifically

The crawler access check is quick if you know which user agents to look for. The main ones to confirm are allowed:

• Googlebot and Bingbot: AI Overviews and Copilot are built on their indexes.
• OAI-SearchBot and ChatGPT-User: how ChatGPT search finds and fetches pages.
• PerplexityBot and Perplexity-User: Perplexity's index and live fetches.
• Claude-SearchBot and Claude-User: Claude's search and user-triggered fetches.

Training crawlers such as GPTBot, ClaudeBot and the Google-Extended token are a separate decision; blocking them does not remove you from search-based answers. Then check your CDN or bot-protection settings, which increasingly block “AI bots” as a category by default.

Re-running the checklist

Tier 1 is not a one-off. Re-check crawler access whenever your hosting, CDN or security settings change, and re-run the entity audit once a year or after any change of name, address, phone number or ownership. Those are the moments inconsistencies creep back in.

Tiers 2 and 3 move at the pace of your team. A reasonable rhythm for a small business is one tier 2 item per fortnight until they are done, then one tier 3 item per month alongside a monthly measurement run.

Frequently asked questions

What should I do first for AI search optimization?

Measure a baseline across twenty buyer questions and three assistants, then audit entity consistency. Those two produce the information every subsequent decision depends on, and entity work is usually the highest-return fix available.

Is publishing more content on the AEO checklist?

Only in tier three, aimed at gaps you have measured. Publishing at volume before identity and corroboration are fixed tends to dilute quality signals without improving citation.

Which AI crawlers should I allow?

For most businesses, the search and user-triggered crawlers: Googlebot, Bingbot, OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, Claude-SearchBot and Claude-User. Training crawlers are a separate choice.

How many prompts should a baseline include?

Twenty is enough for most small and mid-sized businesses. It is small enough to run monthly and large enough to show a pattern across recommendation, comparison and problem questions.