Your buyer asked a machine instead of asking Google, the machine read one page of your website, and the answer it needed was sitting three clicks deep where nothing went looking. Here is why that happens and what to change first.
The visitor you are not designing for
An in-house lawyer at Anthropic described how the job changed. Not "AI writes my briefs now." Something stranger. The work had shifted toward doing "SEO for legal advice, you need to write frameworks that [clients'] agents discover." That is from Anthropic's own employment study, published October 16, 2026.
A lawyer at a frontier AI company restructured how they write so other people's software could find it.
If that sounds like a tech company problem, consider what shipped the same week. On Monday, October 19, 2026, the Superhuman AI newsletter led with ChatGPT learns to use Chrome. Not reading the web. Driving the browser, seeing what is on screen, taking action across tabs.
So, the direct answer to the title question. AI probably is not recommending your business because the specifics a buyer needs, what you charge, who you serve, how the work goes, and your proof, are not in plain text on the page the machine is most likely to read. Your homepage is a mood board. The substance sits on interior pages, behind a form, or inside a PDF.
Most small business websites are built backwards for a machine reader, and the fix is cheap. I built a digital marketing agency before I sold it, so I have strong opinions about site structure. The strongest: almost nobody needs a rebuild. They need three paragraphs moved to the front page.
Key Takeaways
Renders as a shaded callout box.
- Your homepage carries weight it was never designed to carry: your first visitor is often a machine that reads one page and leaves.
- Similarweb reports roughly six in ten ChatGPT referrals now land on a homepage, up from under three in ten before May 2026.
- Google says no special AI files or schema are required, only that important content be in text.
- This evidence is directional, not settled, and certainty about AI crawler behavior is not for sale.
- The cheap fix: answer the three questions a buyer asks before buying, in plain words, up front.
The problem is not that you fell behind
You built a website that works. It loads fast, it looks credible, and the buyer's journey moves from interest to inquiry the way good marketers were taught to build it. Broad promise at the top. Detail as the visitor commits. Pricing after a conversation, because pricing without context kills deals.
That logic is still correct for humans. The reader changed.
This is where I watch capable people come apart, and I want to name it, because the hype machine profits from your panic. Experienced professionals are not overwhelmed by AI. They are quietly convinced that thirty years of hard-won judgment just became worthless. That is confidence collapse, and it is a lie. Your instincts about what a buyer needs to hear are the most valuable asset in this conversation. The delivery mechanism changed. The judgment did not.
Here is the mechanical problem.
A human who lands on a vague homepage will click. They tolerate ambiguity for about four seconds, then hunt, assembling a picture of you from five pages. Your site was designed for that assembly process.
A machine reading your site on a buyer's behalf behaves differently, in ways nobody outside these companies can fully document. Sometimes it crawls broadly. Sometimes it fetches one page and answers from what it found. Sometimes it is a browser agent on your homepage, reading rendered text, deciding whether you are worth recommending. None of those reliably does the assembly job.
So the buyer asks, "who should I hire for this in Grand Rapids, and what does it cost," and your competitor gets named because their front page states what they do, who it is for, and a price range. Yours says "partnering with you to unlock extraordinary outcomes."
You did not fall behind. You optimized for a reader that is no longer the only one showing up.
What the evidence actually shows, and where it gets thin
I am going to separate the solid from the directional, because this category is full of confident numbers that fall apart when you ask what they counted.
Solid: fewer clicks when a summary appears. Pew Research Center analyzed the browsing of 900 US adults across 68,879 Google searches in March 2025. Users who saw an AI summary clicked a traditional result in 8% of visits. Users who did not clicked in 15%. Clicks inside the summary happened in about 1% of visits, and users ended their session entirely on 26% of summary pages against 16% without.
One more Pew figure that should change how you write: 60% of searches beginning with a question word produced an AI summary. Your buyers ask in sentences now.
Solid, and the most useful number here: the homepage is where they land. Similarweb reports that after ChatGPT changed how it surfaces brand links in May 2026, homepage referrals rose from roughly 26% of its referral traffic to about 62% by late May, and held rather than fading. Their read: "roughly six in ten referred visits land on a homepage rather than a specific page." Someone landing there from inside an AI answer has often already made up their mind.
Solid: Google says there is no secret file. Google's documentation on AI features states there are no additional requirements to appear in AI Overviews or AI Mode, and that "you don't need to create new machine readable files, AI text files, or markup to appear in these features." What Google does list includes internal links, structured data matching the visible text, and this, which is the whole article in ten words: "Making sure that important content is available in textual form."
Directional, so label it that way: the crawler ratios. You have seen the claim that AI companies crawl thousands of pages per visitor they send. Duane Forrester documented in Search Engine Journal that Anthropic's crawl-to-refer ratio has been published as 70,900 to one, 38,000 to one, 23,951 to one, 10,300 to one, 4,580 to one and 2,237 to one, among others. Every figure attributed to Cloudflare, all within about thirteen months. Cloudflare itself disclosed that referrals only count when a Referer header names the platform, that Claude's native app sends none, and that the ratios may therefore overstate the imbalance by an amount that is, in its words, "unclear by how much."
The direction is real. The precision is theater. Do not rebuild your business around a number nobody can define.
One practitioner's observation, not research. Andy Crestodina of Orbit Media reported on X that he reviewed Cloudflare AI crawl reports across 74 websites and found AI crawlers hitting homepages far more heavily than other pages. Labeling that accurately: a credible practitioner's observation on social media, date unconfirmed, never published as a study. Treat it as a reason to check your own logs, not as a finding.
One last note from that same Superhuman issue: attackers are already hiding prompt injections in content agents read and humans cannot see. These readers are not careful.
The fix: put the answer on the front page
AI amplifies human skill. It does not replace it. I use the conductor and the orchestra as my frame, and it holds up here better than anywhere. The orchestra got enormous. It plays whatever you put in front of it, loudly and without judgment. What it cannot do is decide what the music should be. That is you. Your thirty years of knowing what a buyer worries about at 11pm is the score. The machines are the section players.
So stop asking how to optimize for the machine. Ask what your buyer needs to know before they will buy, then make that answer readable by anything that opens your front door.
Three questions. Nearly every buyer asks them, in some form, before money moves.
-
Do you do the specific thing I need, for someone like me? Not your category. The specific thing. "We help independent insurance agencies under twenty staff replace manual quoting" beats "insurance technology solutions" for the human and the machine alike.
-
What does it cost, roughly? This is where most owners fight me. You do not need a rate card. You need a range with context. "Most engagements run $4,000 to $12,000 depending on team size" qualifies better than a form does and gives a machine something concrete to cite. A price range is not a surrender. It is a filter.
-
How do I know you can actually do it? Named proof in text. Not a logo wall, which is an image and reads as nothing. Not a slider, which may never render. A sentence a reader can quote.
Put those answers in plain HTML text, above the point where a visitor would scroll twice. Write each heading as the question your buyer asks, then answer it in the first sentence underneath, before any build-up. Answer first, elaborate second. It serves the human reader identically.
Then the supporting moves. Add Organization, LocalBusiness and FAQPage structured data where it fits, which removes ambiguity rather than acting as a visibility lever. Skip llms.txt; Google lists it among tactics owners can ignore. And if your key claims live inside late-rendering JavaScript, an image, or a PDF, assume they do not exist.
Six things to do this week
1. Read your homepage as a machine would. Disable JavaScript, or view the page source, and write down every factual claim that survives. If your price range, specialty and proof are not on that list, you found your problem.
2. Pull the three questions out of your own sales history. You already know them; you have answered them on hundreds of calls. Get them out of your head with this prompt.
[The Job]
Identify the three questions my buyers ask before they commit.
This is for: [MY IDEAL CLIENT, in one sentence].
It matters because: these answers need to sit in plain text on my homepage, where buyers and AI assistants find them without clicking deeper.
[The Background]
Here is what you need to know: I sell [SERVICE] to [AUDIENCE] in [MARKET]. Typical engagement is [PRICE RANGE]. My differentiator is [DIFFERENTIATOR]. My most common objections are [OBJECTION 1], [OBJECTION 2], [OBJECTION 3]. Notes from recent sales calls: [PASTE NOTES].
Do not use: marketing language, superlatives, or any claim I have not evidenced.
[The Deliverable]
Return: the three questions phrased as a buyer says them out loud, each with a two-sentence plain answer drawn only from what I gave you. Under 300 words.
Must include: a flag on any answer where my input was too thin to answer honestly.
Optimize for: accuracy.
[The Questions]
Ask me any questions you have.
3. Write the front-page answer block. Three question headings, each answered in the first sentence, high on the page, in text.
[The Job]
Draft a homepage answer block stating what I do, who it is for, what it costs and why I am credible.
This is for: [AUDIENCE] comparing providers, possibly by asking an AI assistant.
It matters because: a machine reader may fetch only this page, so the specifics cannot sit three clicks deep.
[The Background]
Here is what you need to know: my three buyer questions and answers are [PASTE FROM STEP 2]. My price range is [RANGE] and the factors moving it are [FACTORS]. My verifiable proof points are [PROOF I CAN STAND BEHIND].
Do not use: em dashes, emojis, superlatives, or any statistic I have not supplied. Do not widen the price range until it means nothing.
[The Deliverable]
Return: three H2 headings phrased as the buyer's own question, each with a 40 to 70 word answer leading with the direct answer. Plain HTML-ready text, no images. Written from the perspective of a practitioner who has done this work speaking to a buyer who has been oversold before.
Must include: a specific price range, a named specialty, one concrete proof point.
Optimize for: accuracy.
[The Questions]
Ask me any questions you have.
4. Surface what is buried, and ungate one asset. Find the interior page answering your buyer's hardest question, lift its three best sentences onto the homepage, and link down to the full page. Then publish the single most useful thing you hide behind a form as a readable page, because a form is a wall to a machine.
5. Set up tracking so you can see it. In GA4, open Traffic acquisition, switch to Session source / medium, and look for chatgpt.com, perplexity.ai, claude.ai, gemini.google.com and copilot.microsoft.com. Build a custom channel group above Referral in the rule order, or those sessions get claimed as generic referrals. Expect undercounting, since much of this traffic arrives with no referrer.
6. Re-read it as a human in two weeks. If the page got clearer for the machine and worse for the person, you overcorrected. Fix the person's experience first; they still sign the check.
Frequently Asked Questions
Should I create an llms.txt file for my website?
Not as a priority. Google's John Mueller has called llms.txt purely speculative and said no AI system was using it, and Google's guidance lists it among tactics owners can ignore. It is harmless and takes minutes, but it is not why you are invisible. Plain text on your real pages matters more.
Does schema markup help my business show up in AI answers?
Structured data still earns rich results in ordinary search, so it is worth having. Google states that no special schema.org markup is required for its AI features, and that markup must match the visible text. Use Organization, LocalBusiness and FAQPage markup to remove ambiguity, not as a shortcut.
How can I tell if AI tools are sending traffic to my website?
In GA4, switch Traffic acquisition to Session source / medium and look for the assistant domains directly. Build a custom channel group above Referral so those sessions are not absorbed as generic referrals. Expect a low count, because much of this traffic carries no referrer and lands in Direct.
Will AI agents fill out my contact form or book a call?
Some can navigate forms, and agentic browsing is moving that way fast, but do not design around it. Treat a form as a barrier that may stop a machine entirely. If what sits behind the gate qualifies a buyer, publish enough in readable text for an agent to carry forward.
Is answer engine optimization different from SEO, and do I need to hire someone?
Mostly it is the same work with better discipline about clarity. Google's position is that foundational SEO practices remain the requirement for its AI features. Before hiring a specialist, do the free version: put your specialty, price range and proof in plain text high on your homepage, then measure.
The front door is the whole store now
Your website was built on an assumption so reasonable nobody questioned it. A visitor who wants more will click.
That assumption is now partly false, and the false part is growing. Some share of the visitors deciding whether to recommend you will read one page and leave. They will not hunt. They will take what is in front of them and move on to whoever said something specific.
I am not telling you the web is over. The evidence is directional. The homepage referral shift and the click-through decline are well documented; the crawler ratios everybody quotes are a mess. Anyone claiming to know exactly how these systems weight your pages is guessing with confidence.
But you do not need certainty to act, because the action is the one good marketing asked of you all along. Say what you do. Say who it is for. Say what it costs. Say why anyone should believe you. Say it early, say it plainly, and say it in words instead of in a slider.
The machines did not create that requirement. They removed your ability to postpone it.
Your buyer is standing at your front door with a question. Answer it there.
About the author
Jonathan Mast is the founder of White Beard Strategies LLC, an AI education, coaching and training company he started in June 2023. Before that he built and exited Valorous Circle, a digital marketing agency he grew to seven figures, where he developed his direct experience in web and digital marketing work. He brings a 30-year background in consultative selling to practical, anti-hype AI education for entrepreneurs, and serves a free Facebook community of more than 500,000 members called AI Prompts for Entrepreneurs. The training and live implementation work happen inside White Beard Strategies AI Insiders(TM).
Sources
- Employment at Anthropic, Anthropic, October 16, 2026
- ChatGPT learns to use Chrome, Zain Kahn, Superhuman AI, October 19, 2026
- Google users are less likely to click on links when an AI summary appears, Athena Chapekis and Anna Lieb, Pew Research Center, July 22, 2025
- AI Search Stats 2026, Maayan Zohar Basteker, Similarweb
- AI features and your website, Google Search Central
- Optimizing for generative AI search, Google Search Central
- Everyone Is Quoting The Same Number And Getting Different Answers, Duane Forrester, Search Engine Journal
- The crawl-to-refer ratio on Cloudflare Radar, Cloudflare
- Google: No AI System Currently Uses LLMs.txt, Barry Schwartz, Search Engine Roundtable
- Andy Crestodina, Orbit Media, reported on X. Practitioner observation, not research.