Contents12 sections
When a buyer asks ChatGPT who to hire for what you do, does your firm come up? Does the assistant describe you correctly, or a competitor? You may not know, and there is no obvious place to look.
Call it AI or Super Intelligence (SI). Your buyers already ask it before they ask you. In G2’s March 2026 buyer survey, 51% of B2B software buyers said they now start their research with an AI chatbot. 69% said that research led them to a different vendor than planned. The 6sense Buyer Experience Report 2025 found that buyers bought from a vendor on their day-one shortlist 85–95% of the time. An assistant that can’t read your site can’t put you on that list, and nobody calls to say they didn’t find you.
This guide explains what “AI-ready” means for a firm like yours, how to tell where you stand, and what to fix first. If you would rather hand the job over, that is my AI-ready website rebuild.
The short version
- AI-ready means four things. Assistants can get in, read your pages, trust you enough to mention you, and send you people who can reach you easily.
- Most of it is ordinary good web practice. Google says its AI features have no special requirements beyond what Search already needs.
- The common failures are simple. A security setting that turns AI crawlers away, or pages that look blank to them.
- Fix in order. Access, then clarity, then prices and replies. Nobody can promise that an assistant will name you, me included.
What does “AI-ready” mean for a business like mine?
It means an assistant can find your site, understand who you are, trust it enough to quote, and send you people who then get in touch. I check those as four layers, in order.
| Layer | What it means | What usually goes wrong |
|---|---|---|
| Found | AI crawlers can reach and read your pages | A security setting blocks them, or pages look blank to them |
| Cited | Assistants trust you enough to quote you | An unclear company name, vague pages, no authors or dates |
| Chosen | Visitors from AI answers get in touch | No prices, slow pages on phones, long forms, slow replies |
| Agent-ready | AI tools can act for a buyer | Forms a machine can’t use |
A crawler is the program an AI company sends to read web pages, like the one Google uses for search. Order matters: a page that can’t be found can’t be quoted.
Does this affect a firm like mine?
If buyers look you up before they get in touch, yes. If all your work comes through referrals, much less.
A fair caution: G2 asked software buyers, not clients of law firms or clinics. Read its numbers as a direction, not a measure of your sector.
Picture a five-person surveying firm whose work all comes from three building contractors it has known for years. A rebuild would be the wrong spend there. Check that the site isn’t blocking anyone, correct the name and phone number everywhere, and stop.
How can I tell where my site stands?
Three browser checks tell you most of it in under an hour, or my free audit runs them in about a minute.
Ask the assistants about yourself (15 minutes)
Ask ChatGPT, Claude, Perplexity and Gemini what your firm does, then who they would suggest for your main service in your city. Are you there, described correctly? Who is named instead? Answers vary between tries, so this is a spot check, not a measurement.
See your site the way most AI crawlers do (2 minutes)
Most AI crawlers read only the page as your server first sends it, before any scripts run. Vercel’s crawler analysis (December 2024) found that OpenAI’s and Anthropic’s crawlers don’t run those scripts.
In Chrome, press Ctrl+U on your homepage (Cmd+Option+U on a Mac) to see that raw version. Search it for a sentence from your headline. If it isn’t there, most AI crawlers can’t see it either. My own old agency site showed crawlers 0 words; here is how I rebuilt it.
From the field. I rebuilt SitShape, my own sofa marketplace, as 564 finished pages in three languages. Crawlers now get complete pages. Server jobs they triggered in a 45-second sample, before and after: 103 → 0.
Check whether the gate is open (10 minutes)
Two settings decide whether AI crawlers get in. One is robots.txt, a short public file of house rules for crawlers, at yourdomain.com/robots.txt. The other sits with your hosting or security service, such as Cloudflare, which can turn crawlers away at the door.
OpenAI says that sites which opt out of its search crawler won’t appear in ChatGPT’s search answers, except as plain links. In July 2025, Cloudflare began asking every new domain at sign-up whether to allow AI crawlers, with blocking as the default. It has changed those defaults since, most recently in September 2026. A setting chosen at sign-up may still decide who reads your site. More in is Cloudflare blocking AI agents on your site?
Or let my audit run the checks (1 minute)
My free AI-readiness audit reads your site the way AI crawlers do and gives you an SI-Readiness Score out of 100, without an email. With a work email at your own domain, the full audit is free too. It puts 10 questions to each of 4 assistants and shows whether you are named, and who is named instead.
Will AI assistants trust my site enough to mention it?
They mention firms they can identify without doubt and quote without effort. Four habits help most.
Be one company, spelled one way
Use one name, address and phone number on your site, LinkedIn, your Google listing and the directories in your field. Ask your developer for structured data too: a few lines of page code, invisible to visitors, that state your name, logo, founder and profiles for machines. Google says this markup helps it tell your organization apart from others.
Put the answer first
In my experience, assistants quote passages, not whole pages, and a mission statement gives them nothing to quote. Do the first 60 words of your main service page say what you do, for whom, and roughly at what price? If not, rewrite that paragraph first.
Give numbers with dates
The GEO study by Aggarwal and colleagues (KDD 2024) tested rewrites such as adding statistics, quotations and sources. The best raised visibility in AI-written answers by up to 40% in the authors’ tests. “1,100+ pages live, live sitemap count, observed 2026-10-05” says more than “hundreds of pages”.
Name the author, date the page
Google’s guidance on helpful content asks whether it is clear who wrote a page. Freshness counts too: in an Ahrefs study of nearly 17 million cited pages (July 2025), the pages AI assistants cited were on average 25.7% fresher than pages in Google’s ordinary results. Review your key pages each quarter with real edits; a new date on old text doesn’t count.
From the field. For Corvane Legal, a German law firm, I built an AI content engine that keeps 1,100+ pages live on the firm’s own domain. Each reworked article is checked claim by claim against a source file. The engine logs 16 AI crawlers by name, so “can AI read us?” is a log line, not a guess.
Do the visitors AI sends turn into enquiries?
Only if they find a price, a fast page and a quick way to reach you. Being mentioned is half the job.
Can buyers see a price?
Gartner reported in March 2026 that 67% of B2B buyers prefer to buy without a sales rep. In my experience, those buyers look for a number first, and an assistant can’t match you to a budget your site never states. A “from” price or a typical range is enough. Mine are on my pricing page.
Is the site fast on a phone?
Run your homepage through Google’s free test at pagespeed.web.dev. Open the Mobile tab and read what real visitors experience, not the lab score. No real-visitor data yet? Then the lab score is a rough guide. Green on all three measures is what you want.
How fast do you reply?
A Harvard Business Review study (March 2011) found that firms contacting a lead within an hour were nearly seven times as likely to qualify it as firms that tried an hour later. It is an old study, so test it on yourself: send your own contact form from your phone and time the reply. Good looks like a short form, a booking link and a reply time you state and keep.
A worked example: what a check like this typically finds
Picture a 12-person accounting firm that advises owner-managed companies. Its five-year-old site looks fine in a browser. An afternoon with the checks above might typically turn up this:
- Found. Someone accepted the hosting service’s default at sign-up, so AI crawlers are turned away. The service pages are assembled by scripts, and the raw page holds a menu and a footer.
- Cited. Three spellings of the firm’s name across its site, LinkedIn and a directory. The main service page opens with “We are passionate about numbers”. No article has an author or a date.
- Chosen. No prices. An eleven-field form, and replies in one or two working days.
What does each fix cost? Opening the gate takes 20 minutes. Aligning the name is an afternoon of admin. Rewriting six opening paragraphs and adding a “from” price is a partner’s morning, mostly spent deciding. Cutting the form to four fields is an hour for the web person.
The empty pages are the real decision: days of developer work if the platform can send complete pages, a rebuild if it can’t.
Who notices? In a case like this, probably nobody inside the firm for weeks. The change is typically quiet, and not guaranteed: assistants start describing the firm correctly, and now and then a new client says they found it that way. Ask every new client how they found you, and count the answers that name an assistant.
What should I fix first?
Fix in layer order, Found first. Many fixes take twenty minutes.
- Open the gate. A same-day fix.
- Get your words into the page. If the raw page is nearly empty, nothing else counts yet.
- Make your identity and answers clear. One name, answer-first service pages, numbers with dates.
- Make it easy to act. A price, a fast page on a phone, a short form, a quick reply.
- Agents last. An extra, never a replacement.
When is a rebuild worth it?
When step 2 fails and your platform can’t be fixed cheaply: patching failed foundations tends to cost more than starting clean. If steps 1 and 2 pass, you need a focused fix list, not a rebuild.
A rebuild has its own risk: old addresses that stop working. My rebuild for Orvelle Dental, a premium clinic, has not launched yet. Checking its old addresses link by link caught 15 that would have ended on an error page.
What I left off on purpose
- Hidden text or pages shown only to bots. Google wants what machines read to match what people see.
- Dashboards that track AI answers before the basics pass. They measure a problem you haven’t fixed yet.
- Paying extra for llms.txt, a summary file for AI models. Ahrefs studied 137,000 sites: 97% of the llms.txt files it found got no requests at all in May 2026.
- Anyone promising that ChatGPT will cite you. Nobody controls that, including me.
What about AI agents that act for buyers?
They are the newest and least settled part: a cheap bet on top of the basics. An agent is an AI tool that does a task for someone, such as asking for a quote.
Two things are worth doing now. Make your enquiry form a normal form with a visible label on every field, not a widget inside a frame. That helps screen readers and agents alike. Then write down the one question a buyer’s assistant would want answered live, such as availability. None? Stop there. If there is one, look at the Model Context Protocol (MCP), an open standard for connecting AI tools to outside data. My own product Vowko gives its members’ AI agents 9 MCP tools. That is how I know what a small, safe setup takes: mostly read-only, with limits and logs.
What you can do this week
About an hour of your own time.
- Ask four assistants about your firm (15 minutes). Save the answers to compare in three months.
- Search the raw page (2 minutes). Ctrl+U on your homepage, then find your headline.
- Ask your web person (10 minutes): do robots.txt and our Cloudflare or hosting settings let AI search crawlers in, by decision or by default?
- Read your first 60 words (10 minutes). What you do, for whom, roughly what it costs?
- Check your name everywhere (20 minutes). Site, LinkedIn, Google listing, the directories in your field.
- Test your enquiry path (5 minutes, then wait). On your phone, on a weekday morning.
- Run the free audit (1 minute) on the audit page, for a baseline.
For your web developer
- Crawler access. In robots.txt, allow OAI-SearchBot, Claude-SearchBot and PerplexityBot; decide on GPTBot and ClaudeBot on purpose (OpenAI, Anthropic, Perplexity). In Cloudflare, since 15 September 2026, setting Training to Block, rather than Disallow AI Training, also blocks multi-purpose crawlers such as Googlebot, Bingbot and Applebot.
- Raw HTML. Headlines, services, prices and contact paths in the server’s first response.
- Sitemap. Honest
lastmoddates (Bing: a key recrawl signal; Google: used only when accurate). Real 404s for unknown URLs. - Entity markup.
Organizationwith logo, founderPersonandsameAs, matching the visible text. - Core Web Vitals on mobile, 75th percentile: LCP ≤ 2.5 s, INP ≤ 200 ms, CLS ≤ 0.1 (web.dev).
- Markdown twins of key pages, via
rel="alternate"ortext/markdown. Cloudflare measured one page at 16,180 tokens as HTML and 3,150 as Markdown. - WebMCP was, in October 2026, a W3C community draft and a Chrome origin trial from version 149. Labelled HTML forms first. llms.txt: generated at build time, if at all.
When should you bring someone in?
Once the gate is open and your words are in the page, most of the rest is writing and decisions you can make in-house. Bring someone in when crawlers see empty pages, or nobody on your side can change the site.
The quickest first step is my free audit. For a page-by-page check with fixed quotes, the AI Visibility Deep Audit is €1,500$1,650. Fixed price. What you pay in cash is credited in full to any build within 60 days. The AI-Ready Website Rebuild is from €6,500from $7,200 over 3–6 weeks.
Tell me about your site and I’ll say which you need, including when the answer is neither.
Sources
Every factual claim above traces to one of these. Numbers about my own work come from my dated receipts.
- G2 via PR Newswire: Half of B2B software buyers now start their research with AI chatbots (April 2026) prnewswire.com
- 6sense: Buyer Experience Report 2025 6sense.com
- Google Search Central: AI features and your website developers.google.com
- OpenAI: Overview of OpenAI crawlers developers.openai.com
- Anthropic: Does Anthropic crawl data from the web, and how can site owners block the crawler? support.claude.com
- Perplexity: Perplexity crawlers docs.perplexity.ai
- Cloudflare press release: Cloudflare just changed how AI crawlers scrape the Internet-at-large (1 July 2025) cloudflare.com
- Cloudflare Blog: Have it both ways: stay discoverable in search while disallowing AI training (15 September 2026) blog.cloudflare.com
- Vercel: The rise of the AI crawler (December 2024) vercel.com
- Bing Webmaster Blog: Keeping content discoverable with sitemaps (July 2025) blogs.bing.com
- Google Search Central: Build and submit a sitemap developers.google.com
- Google Search Central: Organization structured data developers.google.com
- Aggarwal et al.: GEO, Generative Engine Optimization (KDD 2024) arxiv.org
- Google Search Central: Creating helpful, reliable, people-first content developers.google.com
- Ahrefs: AI assistants prefer to cite fresher content (July 2025) ahrefs.com
- Gartner: Sales survey finds 67% of B2B buyers prefer a rep-free experience (9 March 2026) gartner.com
- web.dev: Web Vitals web.dev
- Harvard Business Review: The Short Life of Online Sales Leads (March 2011) hbr.org
- W3C Web Machine Learning Community Group: WebMCP, Draft Community Group Report webmachinelearning.github.io
- Chrome for Developers: WebMCP developer.chrome.com
- Model Context Protocol: What is MCP? modelcontextprotocol.io
- Cloudflare: Introducing Markdown for Agents (February 2026) blog.cloudflare.com
- Ahrefs: We analyzed 137K sites, 97% of llms.txt files never get read (June 2026) ahrefs.com
Changelog
- First published.
Reviewed by Vegard Wikeby on . The Updated date changes only when the substance changes, never for a typo.