AI search settings checklist (and what Google says to skip)
Three North Alabama businesses, three different websites: a Decatur family practice on WordPress, a Hartselle hometown pharmacy on Wix and a Decatur equipment rental company on Wix. In October 2026, we read the settings behind every page on all three, 110 pages in all. None of the three blocks the AI crawlers. All three have an llms.txt file. And their results over the past year look nothing alike, which is the first lesson of this post: the AI settings are the entry fee, not the strategy.
The pressure to "optimize for AI" is real. Google said at I/O on May 20, 2025 that AI Overviews had more than 1.5 billion monthly users. Pew Research Center, studying the browsing of 900 U.S. adults in March 2025 (published July 22, 2025), found people clicked a traditional result on 8% of visits when an AI summary appeared, versus 15% when it didn't. So it is no surprise that a cottage industry of checklists, plugins and paid services has appeared.
The problem is that much of that advice is noise, and some of it contradicts what Google says in its own documentation. So this checklist does two jobs. It tells you what is worth checking, and it tells you plainly what Google says you don't need. It is part of our Search Everywhere field study, an observation of three DIWYM clients, not an experiment.
Start with what Google says you don't need
Google's own page, "AI Features and Your Website" (Google Search Central, last updated December 10, 2025), says there is no special markup or file required to appear in AI Overviews or AI Mode. Put that next to the checklist and the picture is simple:
Setting | Required by Google for AI Overviews or AI Mode? | Our view |
Special "AI schema" | No | Nothing special is required |
An llms.txt file | No | A low-cost courtesy, not a ranking factor |
FAQ markup | No | Useful for stating answers precisely, not for a special result |
Structured data that matches the page | Listed as a best practice | Do this one |
Not blocking crawlers | Crawlers must be able to reach your pages | Check it |
Google's guidance also lists "making sure your structured data matches the visible text on the page" among its best practices for AI features. Keep that sentence in mind. It is the thread that runs through the whole checklist.
1. Crawler access: make sure you aren't locking the door by accident
AI assistants can only read and cite what their crawlers can reach. A crawler is an automated visitor, and your site gives each one instructions in a plain text file called robots.txt. Some visitors are named in that file: OpenAI's GPTBot, Anthropic's ClaudeBot, Perplexity's PerplexityBot, and Google's own AI crawlers.
What we found: in the October 15, 2026 audit, 0 of 3 sites block any of them. All three are open.
Why this is worth a check: blocking is easy to do by accident. A security plugin, a host's default setting or an old rule left behind can shut the door without anyone deciding to. If you want AI assistants to be able to read your site, make sure no rule says otherwise. (Blocking is also a legitimate choice for some businesses, such as publishers who don't want their writing used. For a local business that wants to be found, it rarely makes sense.)
An honest note on what "open" buys. Open doesn't mean cited. The pharmacy's website analytics have labeled five visits as coming from AI platforms since the site launched in May 2024, through October 14, 2026, and the rental company's have labeled none. That is a handful. Some AI-assisted discovery may be invisible to these tools, which is a story for our post on "direct" traffic. The point of crawler access is not a traffic spike. It is simply not being locked out.
Check it today: type your web address followed by /robots.txt into a browser. Look for lines that say Disallow: / under a bot's name. If you see one you didn't intend, ask whoever manages your site to remove it.
2. llms.txt: a courtesy, not a requirement
An llms.txt file is a plain text summary of your website, written for AI tools: who you are, what the main pages are, where to find them. Think of it as a cover sheet.
What we found: all three sites have one. The practice's is generated by its SEO plugin, and the Wix sites' are generated by Wix.
What Google says: it is not needed for AI Overviews or AI Mode. We agree, and we treat the file as a low-cost courtesy for other AI tools, not a ranking factor. If your site builder or plugin makes one automatically, leave it on. If someone offers to sell you an "AI visibility file" for real money, they are selling something Google says you don't need.
Check it today: open yourwebsite.com/llms.txt. If it loads, you have one. If it doesn't, don't lose sleep.
3. Page titles and meta descriptions: the short summaries everyone reads
A page title is the headline of a search result. A meta description is the one or two sentence summary under it. Search engines and AI tools often repeat or draw on both, though Google may rewrite them to suit a search.
What we found: 106 of the 110 pages have a meta description: 76 of 80 on the practice's site, 18 of 18 on the pharmacy's and 12 of 12 on the rental company's. The four without are blog category listing pages, not service or location pages.
Why they matter: they are the cheapest, clearest statement of what a page is about. In our read, an AI tool deciding what a page covers uses the same signals a person scanning results would.
What a good one looks like:
Start with what the page is about, in the words a customer would use. For a service page, that is the service and the town.
Make each one unique. Two pages with the same title confuse people and machines alike.
Keep titles to roughly 50 to 60 characters and descriptions to a couple of short sentences, so they aren't cut off. That is our working guideline, not a Google rule.
Say something true that the page delivers. Don't promise what the page doesn't contain.
Check it today: search your business and service on Google. Read the title and description under your result. Do they say what you would want a stranger to read?
4. FAQ markup: what changed in 2023, and why we still use it
FAQ markup is structured data that labels questions and answers on a page, so a machine can read them cleanly. Years ago, it could earn a business an expanded drop-down result in Google. On August 8, 2023, Google limited those FAQ rich results to well-known, authoritative government and health websites (Google Search Central Blog, "Changes to HowTo and FAQ rich results"). For nearly every other business, the special format is gone.
So why do we still add it? Not to win a special format. We add it to state answers precisely, so the page tells search engines and AI tools in an unambiguous way what a customer's question is and what the answer is. We don't count on any visible reward for it, and we can't show that it produced one.
What we found: 63 pages carry FAQ markup: 62 on the practice's site and one on the pharmacy's, a page with twelve marked-up questions. The annual physical service page on the practice's site marks up eight. The rental company has none.
The rules we follow:
Answer real questions. Use questions customers actually ask, taken from calls, the front desk and the counter.
Match the visible text. Every answer in the code must be words a visitor can read on that page. This is what Google's own AI features guidance says about structured data in general. Never put in the code something the page doesn't say.
Keep answers specific and short. One clear statement, not a sales pitch.
Update both together. When the answer changes on the page, change it in the markup the same day.
Mark up only what exists. No hidden questions, no stuffing the page with questions for the sake of it.
Here is the question and answer part of a block from the practice's annual physical page, trimmed and anonymized, so you can see what "matching the visible text" looks like. The real page marks up eight questions; two are shown.
{
"@context": "https://schema.org",
"@graph": [
{
"@type": "FAQPage",
"@id": "https://www.example.com/services/annual-physical/#faq",
"mainEntity": [
{
"@type": "Question",
"name": "What is included in an annual physical exam?",
"acceptedAnswer": {
"@type": "Answer",
"text": "An annual physical typically includes a review of your health history and current medications, a check of your vital signs such as blood pressure and weight, and a general wellness assessment."
}
},
{
"@type": "Question",
"name": "How do I schedule an annual physical?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Call us at (256) 555-0123 to schedule your visit."
}
}
]
}
]
}Each question has a name, which is the question as a visitor reads it, and an acceptedAnswer with a text, the answer. The phone number in the second answer is the same one printed on the page. If the page changed and the code didn't, they would disagree, and that is exactly what to avoid.
For a line-by-line walk through the full block and the business record around it, see Rich schema in plain English.
5. Structured data that matches the page and the profile
The one structured data rule Google does name for AI features is that it should match what visitors see. Take it one step further and make it match your Business Profile too: name, address, phone and hours, character for character. In our October 2026 comparison, the practice's and both pharmacy stores' name, address, phone and hours matched their profiles on the day checked. The rental company's name, address and phone matched; its hours aren't in its markup. For two-store hours, see Two locations, two sets of hours.
What the audit looked like
Practice (WordPress) | Pharmacy (Wix) | Rental company (Wix) | |
Pages audited | 80 | 18 | 12 |
AI crawlers blocked | None | None | None |
llms.txt | Yes (SEO plugin) | Yes (Wix) | Yes (Wix) |
Meta descriptions | 76 of 80 | 18 of 18 | 12 of 12 |
FAQ-marked pages | 62 | 1 | 0 |
What this checklist doesn't prove
These three sites share the same AI access settings (open crawlers, an llms.txt file) and yet their results differ, so those settings did not decide the outcome. From July 23 to October 14, 2026, compared with the same dates in 2025, the rental company's Google impressions rose 107% (8,148 to 16,873) and its clicks 16% (886 to 1,024). The practice's impressions rose 36% (26,782 to 36,377) while its clicks were flat, down 2% (2,154 to 2,120). Those results came with other work: new pages, clear structure, Business Profile care. We can't separate what any one setting contributed, and we don't claim to.
What we can say: these settings are cheap, reasonable and worth getting right, and none of them is a shortcut. A site that is blocked, unclear or contradicting its own profile has a problem no AI setting will solve, and a site that is clear and consistent has less to fix.
Your five-minute version
Open /robots.txt and confirm no rule blocks AI crawlers you want to read your site.
Open /llms.txt and see whether one exists. Leave it alone if it does.
Search your business and read the title and description under your result.
Run Google's Rich Results Test on a page with FAQs and compare each answer to the page.
Compare your name, phone and hours on your site with your Business Profile.
If a service is helpful here, our local SEO and AI enhancements cover the setup, and our website development team can handle the pages themselves. For more context on how Google builds those summaries, see The anatomy of an AI Overview.
Frequently asked questions
Do I need an llms.txt file to show up in AI answers?
No. Google says no special markup or file is required to appear in AI Overviews or AI Mode, and that includes llms.txt. All three sites in our study have one, generated by an SEO plugin or by Wix, and we treat it as a low-cost courtesy for other AI tools, not a ranking factor. If your site makes one automatically, keep it; don't pay extra for one.
Should I block AI crawlers like GPTBot or ClaudeBot?
It depends on your goals. Publishers who don't want their writing used may choose to. A local business that wants to be found usually shouldn't. In our October 15, 2026 audit, none of the three sites blocked GPTBot, ClaudeBot, PerplexityBot or Google's AI crawlers. Check your robots.txt to be sure a plugin or host setting isn't blocking them by accident.
Does FAQ markup still work after Google's 2023 change?
Not for special results, in most cases, and not for most local businesses. On August 8, 2023, Google limited FAQ rich results to well-known, authoritative government and health websites. We still add FAQ markup, to 63 pages across two of our client sites, because it states answers precisely for search and AI tools. We don't expect a special format, and we can't show a measured benefit.
What does "structured data must match the visible text" mean?
Every answer, hour and fact in your code should be readable on the page itself. Google lists this among its best practices for AI features. If the markup says a store closes at 5 p.m., the page and your Business Profile should say 5 p.m. too. Hidden or contradicting code is a risk, not a bonus. When content changes, update the page and the code together.
Do I need a meta description on every page?
It's a good habit, and an easy one to audit. In our audit, 106 of 110 pages across three sites had one, and the four without were blog category listings. A meta description is the short summary under your search result, and search engines and AI tools often draw on it, though Google may rewrite it. Make each one unique and describe what the page really offers.
Will these settings increase my AI visibility or clicks?
We can't promise that. The three sites we audited share the same AI settings and had different results from July 23 to October 14, 2026 against 2025: the rental company's impressions rose 107% and its clicks 16%, while the practice's impressions rose 36% and its clicks were flat. The settings help AI tools read your site; they are not a shortcut to traffic.
How often should I run through this checklist?
A few times a year, and any time you change your website platform, add a security plugin or redesign. Settings drift. A plugin update can add a crawler block, a new page template can drop the meta descriptions, and an FAQ can fall out of step with its markup. The five-minute version above is enough for a routine check, and it costs nothing.
Keep reading
What three North Alabama businesses taught us about search in the AI era: the full study, with every trend and its limits
Rich schema in plain English: the code behind these settings, line by line
The AI search timeline every local business owner should know: when each change landed
"Direct" traffic is rising: where AI and app visits may be hiding
Schema markup 101: our earlier practical guide
Google AI Overviews: impact on your small business: what the summaries mean for local owners
You don't need a secret file or a special tag. You need a site that is open, clear and consistent with your profile. Call (256) 274-1289 or message us for a free visibility check.