We argued in the pillar piece that the new front door to your business is being the answer an AI assistant gives, and that getting there rewards the same fundamentals as a well-built website. This is the working checklist version: thirteen concrete things to fix, why each one moves the needle, and how to check it yourself.
Print it, run it against your site, and fix the ones you fail. None of it is exotic; all of it compounds.
Structure: can a machine understand what you are?
1. Add Organization / LocalBusiness JSON-LD.
Why: It states your name, address, hours, phone, and category in a format machines read without guessing, and it feeds rich results in classic Search. Check: paste your URL into Google’s Rich Results Test. If nothing shows, you have none. Corrected August 21: Google’s official guidance, updated July 10 2026, says structured data is not required for generative AI search and that there is no special schema markup to add for it. Ship it anyway for rich results, for the non-Google systems that do read it, and because it forces you to state your facts consistently. Just do not buy it as an AI-visibility product.
2. Add FAQPage structured data to your FAQ.
Why: Same qualification as item 1: useful markup, not an AI lever. The thing that actually gets quoted is the plain question-and-answer text on the page, which is item 5 below. Check: same Rich Results Test; look for FAQ items.
3. Use one real <h1> and a logical heading order.
Why: Headings are how a parser maps your page. A wall of styled <div>s tells it nothing. Check: view the page outline; it should read like a table of contents.
4. Give every meaningful image descriptive alt text.
Why: It’s content a machine can read, and it’s an accessibility requirement either way. Check: no empty alt="" on anything informative.
Content: do you answer the questions people ask?
5. Have a page (or section) per real customer question. Why: Assistants extract answers to specific questions. “How much does X cost,” “do you serve Y area,” “how long does Z take” should each have a plain answer somewhere. Check: list the five questions a customer asks before hiring you. Can you point to where each is answered?
6. Answer concisely before you elaborate. Why: A clear one-or-two-sentence answer up top is what gets quoted; the supporting detail can follow. Check: could someone copy your first two sentences as the answer?
7. Be specific and honest about pricing and scope. Why: Vague “contact us for pricing” gives an assistant nothing, and “starts at $X” is far more likely to be surfaced. Check: is there a real number or range anywhere?
8. Keep your facts current. Why: Stale hours, old services, or last year’s team erode the trust these systems place in your site. Check: when did you last update the about and services pages?
Performance & accessibility: can it be read at all?
9. Render content on the server (or pre-render it). Why: If the real content only appears after JavaScript runs, crawlers and agents often get a blank page. Check: disable JavaScript and reload. Is your content still there?
10. Pass Core Web Vitals, especially INP. Why: Slow, janky pages get crawled less and rank worse, in classic search and AI answers alike. Check: run PageSpeed Insights; aim for green.
11. Meet basic accessibility (WCAG). Why: Semantic, accessible markup is, almost by definition, machine-readable markup. Headings that describe structure, real link text, and alt attributes are what let an assistant work out what a page says. In Ontario it is also a legal requirement for organizations with 50 or more employees and for the public sector, though not for smaller businesses, which does not change the fact that it helps every site. Check: run a free axe or Lighthouse accessibility scan.
12. Confirm nothing at your edge is blocking the crawlers you want.
Why: Every item above assumes the crawler gets a response. A CDN or firewall rule that refuses AI crawlers, or a robots.txt line somebody added in 2024, quietly undoes all of it, and it fails silently. This got sharper in 2026: Cloudflare is retiring its single “Block AI bots” toggle on September 15 and classifying Googlebot, Bingbot, and Applebot as multi-purpose crawlers, which means any setting that blocks AI training now blocks them too. Check: run a live URL inspection in Google Search Console, and read your robots.txt and your CDN’s bot settings. Full write-up of the Cloudflare change and what to set.
Trust & consistency: do the signals agree?
13. Make your name, address, and phone identical everywhere. Why: If your site, Google Business Profile, and old directory listings disagree, an assistant can’t tell which is right, so it may pick wrong or leave you out. Check: search your business name and compare the details across the top results. (This overlaps heavily with classic local SEO.)
The five-minute audit
Once you’ve worked the list, test it the way a customer would:
- Ask Google’s AI Mode and ChatGPT: “best [your service] in [your city]” and “is [your business] any good.”
- Ask an assistant to summarize your homepage.
- If you’re missing, misdescribed, or it can’t parse your site, go back to the item above that explains why.
Added August 21: there is now a real measurement, and a switch that can undo all of this. Search Console has a generative AI performance report showing how often your pages appear in AI Overviews and AI Mode. It reports impressions only, with no clicks and no click-through rate, so read it as a visibility signal rather than a traffic number. If you looked for it earlier in the year and did not have it, look again: Google began the rollout on June 3, 2026 to a subset of properties and finished pushing it to every Search Console property on August 11, without a formal announcement. It is under Performance, either as its own entry or nested inside Search results depending on the account, which is why people who have it still report not being able to find it. Search Console also now carries a setting under Settings, called Search generative AI, that removes your site from those features entirely. Include is the default, but anyone with owner access to your property can change it, which is a good reason to audit who still has access. Full explanation of both.
Added September 13: a second thing that is not on this list, and it is the one being sold hardest right now. If someone tells you your site needs WebMCP tools to be found by AI, they have the wrong end of it. WebMCP is a draft standard that lets a page hand an AI agent callable functions, and it only does anything once an agent is already on your page. It has no effect on ranking, citation, or whether an assistant recommends you, which is the entire subject of this checklist. It is also still an experiment: it is in a Chrome origin trial that expires at Chrome 156 on October 20, Chrome Platform Status lists the feature as Proposed with no committed ship date, WebKit’s public position on it is oppose and Mozilla’s is neutral. Shopify merchants are the one group with a reason to care this month, because Shopify switched ten of these tools on across every Liquid storefront in August without asking, and one of them reads your policy pages out to agents. That is a customer service issue rather than a discovery one, and we covered it in the post on Shopify’s WebMCP tools.
One thing that is not on this list, deliberately: an llms.txt file. Google’s July 10 2026 guidance names it specifically among the machine-readable files you do not need for its generative AI features. We publish one on this site because it is cheap and other systems may use it, but it does not belong on a checklist of things that decide whether you get recommended.
That’s the whole game: there’s no trick, just a well-structured, fast, honest site that a machine can confidently read and recommend. If you’d rather we run this audit on your site and fix what’s failing, send us the URL. It’s the kind of thing we do every week.
