AI search optimization: what to do, in order

What is AI search optimization?

AI search optimization means getting a website into the answers that AI systems write. The systems differ, and it helps to know where each one gets its pages before you plan anything.

The terms GEO, AEO and AIO all describe this work with different emphasis, as our comparison of GEO, SEO and AEO explains. Our 24-point GEO audit is the full checklist. What follows is the order of work: what to do first, who does it, and why that order holds.

The order runs from what can shut you out completely to what merely improves your odds. A blocked bot makes a well-written page worthless, while a slightly weak page can still be used. So there is no reason to rewrite anything until the first three steps are clean.

Step 1: can the engines reach your pages?

This is usually the shortest step and has the most severe failure. Three things to check, in an hour or two.

robots.txt. Open yourdomain.com/robots.txt. Check that OAI-SearchBot, Claude-SearchBot, PerplexityBot and Bingbot are not disallowed, directly or through “User-agent: *”. Blocking training bots such as GPTBot and ClaudeBot is a separate choice, and it does not keep you out of search answers. Our list of AI crawlers has the exact lines.

The firewall. Security services can refuse bots that robots.txt allows. OpenAI’s help page specifically says to confirm that your host or CDN allows traffic from its published searchbot IP addresses, and Perplexity gives firewall instructions for its bots. When we tested the MWY checker on 36 business websites in seven countries on November 3, 2026 (a hand-picked sample, not a market figure), 4 refused the checker itself. Of the 32 we could read, 4 refused a request sent under AI bot names, against only 1 that blocked such a bot in robots.txt. So the firewall is where to look, even if robots.txt is clean.

Google’s own switch. Since August 31, 2026, every site in Search Console has a Search generative AI control under Settings. “Include” is the default, so nothing is likely to be wrong, but look once. If someone has set it to exclude, your links and content stay out of AI Overviews, AI Mode and generative features in Discover, and nothing else on this page will change that.

Step 2: can they read what is on the page?

Once a bot is through the door, it needs text. Google says it can process JavaScript as long as it is not blocked, but Vercel’s December 2024 analysis of its network found that the major AI crawlers it studied did not render JavaScript. Plan for the stricter case.

• Text in the HTML. View the source of your main pages and search for a sentence of your own. If it is not there, the text is built by scripts and may be invisible to some bots. In the same 32 sites, 6 had too little readable text in the raw HTML of the home page.
• A fast response. Servers that take several seconds to answer risk timing out. Fix hosting and caching before anything cosmetic.
• Pages allowed in the index. Search your pages for “noindex” and check in Search Console and Bing Webmaster Tools that main pages are indexed. Google says a page must be indexed and eligible for a snippet to appear in its AI features.
• A sitemap. List your public pages in sitemap.xml and name it in robots.txt.
• Key facts in plain text. Microsoft warns against key information only in images, in PDFs, or tucked into tabs and expandable menus.

A developer or your host can do most of this, usually in a day. On a site built with a JavaScript framework, serving HTML from the server can take longer, and it is often the single most valuable change on the list.

Step 3: can they tell who you are?

An assistant that finds your page still has to link it to a real business. This step is cheap, and the sample above shows how often it is unfinished: only 8 of the 32 sites passed our check for address and phone written as plain text, and only 11 had complete business markup.

• One name, written one way. Use the same form in page titles, schema, the footer and your profiles. Variants can read as different businesses.
• Contact details as text. Address, phone and email in HTML, on the contact page and in the footer.
• Business schema. Organization or LocalBusiness on the home page, with name, logo, contact details and “sameAs” links. Our guide to schema markup has blocks to copy. Google says no special schema is needed for its AI features, so see this as stating facts clearly, not as a ranking trick.
• An About page. Say who runs the business, since when, and where.
• Profiles that link back. Google Business Profile for anything local, plus LinkedIn or whichever networks you really use, linked from your site.

Step 4: do your pages answer real questions?

This is the longest step and the one that lasts. Google writes that creating content people find unique, compelling and useful “will likely influence your website’s presence in generative AI search in the long run more than any of the other suggestions” in its guide.

Start from questions you actually get. Your inbox, sales calls and reviews hold them. “How much?”, “How long?”, “Do you serve my area?”, “What if it goes wrong?” Keyword tools help with wording but will not tell you what people hesitate about.

Write what only you can write. Google separates commodity content, such as a generic “7 Tips” list, from content with real experience behind it. A roofer who explains why one material fails on north-facing slopes in their climate is far harder to replace than one who repeats general advice.

Put the answer first. Open each section with the direct answer, then explain. State prices, ranges, limits and timeframes wherever you honestly can.

One topic, one good page. Google warns that creating separate content for every variation people might search, primarily to manipulate rankings or AI responses, violates its scaled content abuse policy, and that it is ineffective as well. Cover a topic properly on one page rather than publishing thin pages for each phrasing.

Step 5: are you mentioned elsewhere?

AI answers draw on what is said about you across the web, not only on your site. Google says its AI features can show what is said about products and services in blogs, videos and forums. It also says that seeking inauthentic mentions is not as helpful as it might seem, and that its spam systems apply.

So this step is ordinary reputation work:

• Listings. Keep your Google Business Profile and the directories in your field complete and accurate. A restaurant should look at OpenTable, Resy and Yelp, which ChatGPT’s help page names for reservations.
• Reviews. Ask real customers. Answer them.
• Being worth quoting. Original numbers, a useful tool or a clear opinion can earn coverage, because writers need sources as well.
• Merchant data. An online shop should keep Merchant Center up to date; Google names it for product details in AI responses.

Nobody can tell you how much each of these counts in each system, because none of the companies publish weights.

Step 6: how do you know if it is working?

Measure from three directions, because no single one is complete.

• Google’s report. The Generative AI performance report in Search Console shows impressions from AI Overviews and AI Mode by page, country, device and date. It has been open to all sites since August 31, 2026, though it may not appear for sites with too few impressions.
• Your own questions. Keep a fixed list of 10 to 20 questions and ask ChatGPT, Gemini, Claude and Perplexity each month. Note who is named and linked, and keep the date and exact wording of each question in one sheet so that rounds can be compared. Answers vary between tries, so look for trends over several rounds. Our guide to AI visibility lays out the routine.
• Referrals. Watch analytics for visits from chatgpt.com, perplexity.ai, gemini.google.com, claude.ai and copilot.microsoft.com.

Do not judge a change after a week. Crawling, indexing and answers all take time to move, and nobody publishes how long.

Who does each step, and how long does it take?

These are rough estimates for a site of 10 to 50 pages, not measured figures. The last column matters most, because the steps do not rest on equal proof.

Steps 1 and 2 are close to fact, because a bot that is refused cannot read the page. Steps 3 to 5 are reasoned from what the companies publish and from limited tests, so treat them as good bets. If you only have an hour this month, spend it on step 1.

What can you safely skip?

A worked example: a 10-page local site

An illustrative plan for a hypothetical plumbing company with 10 pages and no developer on staff:

• Hour one, the owner. Reads robots.txt, asks the host about bot blocking, checks the Search Console control.
• Day one, the web person. Removes the security rule that refused AI bots, confirms text is in the HTML, adds a sitemap line.
• Week one, the web person. Adds LocalBusiness schema, puts the address and phone in the footer as text, makes the name consistent.
• Weeks two to four, the owner. Writes down the 10 questions customers ask most. Adds price ranges and response times to service pages, each answer in plain words.
• Each month after that, the owner. Asks real customers for reviews, runs the fixed list of questions, and adds one page when a new question keeps coming up.

Most of the cost here is the owner’s time, and none of it is wasted if AI search moves on, since better pages also help people and classic search.

Where the MWY checker fits

The MWY checker covers steps 1 to 4 from the outside: 24 items on access, identity and content, read from your raw HTML as an AI search bot would see it. The free check gives your score and the five items that matter most, which is a quick way to find where in this order you stand. Pro shows all 24 with the fix for your platform.

Step 5 is outside its reach, since it does not read reviews or other sites. For step 6, the free check does not ask the assistants about you; Pro asks them 20 buyer questions each week. Neither can see inside any AI company’s index. A clean report means the blocks on your side are gone. It does not promise that any assistant will name you.