Google and ChatGPT now answer many questions before anyone visits your site. Using this autumn's data: which searches AI absorbs, which still lead to an inquiry, and what to change on a service business website this month.
- Published
- Reading
- 13 min
- Author
- Yaroslav Tsarenko

A couple of years ago the path to a customer was simple: a Google search, ten blue links, a visit, a call. In autumn 2026 there's an AI layer between the search and your website. Google shows AI Overviews and AI Mode, people ask ChatGPT, and a lot of them get their answer without opening a single site.
This article is for owners of service businesses, agencies and companies whose website is supposed to bring inquiries, not just exist. Using fresh data, we'll look at what actually changed, which searches AI absorbs and which still lead to an inquiry, and what to rework on your site so that AI cites it and people contact you from it.
What changed in search this autumn
In short: fewer clicks go to websites, and more answers end without a click. Here are the facts worth building on.
- Clicks roughly halve when an AI answer appears. Pew Research Center tracked the real Google searches of 900 US adults in March 2025. When Google showed an AI summary, people clicked a regular result in 8% of visits; without one, in 15%. They clicked a link inside the summary itself in just 1% of visits. And 26% of those visits ended the browsing session entirely, versus 16% without a summary (Pew Research).
- An experiment points the same way. In August 2026, researchers from several universities published a preregistered field experiment with 1,100 participants: when AI Overviews and AI Mode were hidden, people clicked through to websites more often (arXiv).
- Google is making links more visible, but it isn't bringing the old results page back. On October 15, 2026, Google said inline links in AI Overviews and AI Mode answers would appear higher and more often, and started testing a separate block of "official pages" for specific tasks (Search Engine Journal).
- There's now separate reporting. Since August 31, 2026, every site in Search Console has a report on impressions in Google's generative AI features. It shows impressions, pages, countries and devices, but not clicks (Google Search Central).
What this means for a business: if you judge your website by visits, the chart will probably keep sliding even when you do everything right. Measure inquiries instead. We'll come back to how.
Which searches AI absorbs, and which still lead to an inquiry
Not all searches are equal. AI is best where a person needs knowledge and weakest where they need to act with a specific company.
| Search type | Example | What AI does | What it means for your site |
|---|---|---|---|
| General knowledge | "how to get a stain out of a couch" | answers it fully | almost nothing: the person is satisfied |
| Comparison and criteria | "deep clean vs regular clean" | explains the difference, cites 3–5 sources | a chance to be a source and be remembered |
| Price and terms | "how much is a deep clean for a 2-bedroom apartment" | gives a range if it can find one on websites | your pricing page is what gets quoted |
| Choosing a provider | "cleaning company near me reviews" | builds a shortlist | you're either on it or you aren't |
| Action | "book a cleaning tomorrow" | sends people to a site or the map | a direct inquiry, if your site is ready for it |
The conclusion is uncomfortable but useful. Blog posts like "10 cleaning tips" lose traffic fastest, because AI retells them in its own words. Pages where people decide and act — prices, services, examples of work, the form — lose the least. Those are where to invest first.
The website's new job: not a reference library, but where the deal closes
A website used to explain, persuade and take the inquiry. Now AI often handles the explaining. People arrive later, with a shortlist already in mind and one question: "Why you, and how soon can you do it?"
- Question"how much does a deep clean cost"
- AI answera range, criteria, 3–5 sources
- Shortlist2–3 companies from the answer and the map
- Service pageprices, examples, terms, reviews
- Inquirya two-field form or a call
Three practical consequences follow:
- The first screen should answer "why you", not "about us". Visitors already know what cleaning or web development is. They need prices, timelines, service area, examples and a way to get in touch.
- Every service deserves its own page. AI breaks a complex question into several sub-queries (Google calls this query fan-out) and looks for an answer to each one. One generic "Our services" page loses to ten specific ones.
- Facts beat adjectives. No model will quote "quality work at fair prices". It will quote "Deep clean for apartments up to 650 sq ft: from your real price, 4–5 hours, two cleaners, supplies included".
Brochure site
- One "Services" pageeverything in a single list
- "Price on request"answers only by phone
- Text rendered in the browserAI crawlers see an empty page
Lead-generating site
- A page per serviceone answer for one query
- Prices or rangessomething AI can quote
- HTML from the serverevery crawler can read the text
How to get into AI answers: what actually works
A whole industry with new acronyms has grown around "optimizing for AI". The primary sources say simpler things.
Ordinary search comes first
Google's documentation is direct: there are no additional requirements to appear in AI Overviews or AI Mode. The page needs to be indexed and eligible to show with a snippet, and you don't need special markup, "AI text files" or other new files (Google Search Central). The same page claims that clicks from results with AI Overviews are higher quality — people who do click tend to spend longer on the site. That's Google's claim, not an independent measurement, but it lines up with the idea that fewer, better-informed visitors arrive.
The data backs up the "ordinary search first" part. On October 29, 2026, Ahrefs published a study in which it asked five AI tools the same questions every day for 20 days and tracked about 37,500 citations. Pages that rank in Google stayed cited in AI Overviews for at least a week in 29% of cases versus 4% for pages that don't rank — roughly seven times as often (Ahrefs). The authors are clear that this is a correlation, not proof of cause, but the direction is plain: good SEO is still the foundation.
AI citations are unstable, so don't panic over one check
From the same study: by the next day, between 50% of citations (AI Overviews) and 83% (Copilot) were gone. At the same time, about 75% of the citations that disappeared came back within the 20 days.
The practical takeaway: don't draw conclusions from one screenshot of "we're not in ChatGPT". Check the same 10–20 questions once a week and watch the trend.
Your text has to be in the HTML, not only in the browser
Googlebot runs JavaScript. Most other AI crawlers don't. Back in late 2024, Vercel analyzed crawler traffic and concluded that the major AI crawlers (GPTBot, ClaudeBot, PerplexityBot and others) don't execute JavaScript; the exceptions were Gemini, which runs on Googlebot's infrastructure, and AppleBot (Vercel).
If your site is a single-page app where the text only appears after scripts load, some AI systems see an empty page. This is common with sites generated in AI builders; I covered other typical problems of such projects in the article on vibe-coded MVPs. The check is simple: open the page, press Ctrl+U (or Cmd+Option+U on a Mac) to view the source, and search for your price or service name. If they aren't there, the problem is real.
Structured data: yes, but honest
schema.org markup isn't a ticket into AI answers, but it helps search engines understand who you are, what you sell and where you work. The key rule from the same Google documentation: structured data must match the visible text. An FAQ in the markup that isn't on the page, or invented reviews, lead to trouble, not citations.
A service page that gets cited and gets inquiries
In 2026 the most valuable page on a service website isn't the homepage but the page for a specific service. That's the page AI quotes for a price or terms question, and the page people land on to decide.
What it needs, top to bottom:
- A headline that names the outcome and the city or format. "Same-day apartment deep cleaning in Austin", not "Our services".
- A price or range and what it depends on. Even when the exact price is custom, a range plus 3–4 factors answers the main question.
- What's included and what isn't. A list that's easy to quote.
- Proof. Real photos, examples of work, reviews with names, case studies with numbers — only real ones.
- Answers to objections. 4–6 questions customers actually ask on the phone.
- A short form. Few fields, no sign-up, spam protection.
- Links to related services and case studies. For people, to compare; for search engines, to understand the site's structure.
A good test: read the page as an AI that has to answer "how much does it cost and how long does it take". If the answer can be written from your page alone, the page is ready.
Forms, speed and trust: where inquiries get lost
With fewer visitors, each one who reaches your site is worth more. Losing them to a long form or a slow mobile load hurts more than it used to.
An example from my own work: for Clean360, a new cleaning company, I built a one-page Webflow site from the client's design. We put prices for 6 core services and 13 add-ons right on the page, showed the real team and reviews, and cut the booking form to two fields — name and phone — protected by Cloudflare Turnstile. According to the client, booking conversion rose 30% in the first two months.
Why it worked, and why it matters even more with AI search:
- Prices on the page answer the visitor and give AI a fact it can quote. "Price on request" helps neither.
- Two fields instead of eight. Someone who already picked you from an AI answer doesn't want to fill in a questionnaire. Details can wait for the call.
- Real team photos. Stock photos don't convince people, and probably not models looking for signs of a real business either.
A word on one page versus many. For a new business with a handful of services, a one-page site is a fine start: fast, cheap, everything in one place. As services and queries grow, separate pages per service give search engines and AI more specific answers to work with. For the AddSale B2B marketplace, the designer and I planned 50 SEO pages in English and Latvian from day one, so the new platform could earn search traffic without waiting for ads. I wrote up what else belongs in the first version of such a product in what to build first in a marketplace MVP.
How to measure when there are fewer clicks
If visits are your main metric, you'll see bad news even while the business grows. Here's a better set.
| What to measure | Where | Why |
|---|---|---|
| Inquiries by the page they were sent from | the form saves the URL to your CRM or a sheet | shows which pages actually sell |
| A "How did you hear about us?" field | in the form, optional | catches ChatGPT, AI Overviews and referrals your analytics can't see |
| Impressions in generative AI features | Search Console, Generative AI report | whether your pages show up in Google's AI answers |
| Visits from ChatGPT and Perplexity | GA4, sources chatgpt.com, perplexity.ai | a trend, not an exact number |
| 10–20 key questions | a manual check once a week | whether you're mentioned in answers to your buyers' questions |
Why analytics alone isn't enough: from October 8 to 15, 2026, links in ChatGPT answers temporarily stopped passing the utm_source=chatgpt.com tag and the referrer, so those visits fell into Direct. According to reports compiled by Search Engine Journal, some sites saw "ChatGPT traffic" drop by about 45% while people kept clicking just the same (Search Engine Journal). A form field and per-page inquiry counts survive glitches like that.
A 30-day checklist
You don't need to rebuild everything at once. This is the order that gets the most out of the least money.
| Week | What to do | How to check |
|---|---|---|
| 1 | View source on the homepage and 2–3 service pages: is the text and pricing in the HTML? | the service name and price are visible in the page source |
| 1 | Add a "How did you hear about us?" field and save the page URL with each inquiry | the first inquiries arrive with this data |
| 1 | Verify the site in Search Console and open the Generative AI report | you see impressions per page |
| 2 | A separate page for each key service: outcome, price or range, what's included | every commercial query has its own page |
| 2 | Cut the form to 2–3 fields and add spam protection | the form can be filled in on a phone in 20 seconds |
| 3 | 4–6 real customer questions on each service page | questions come from actual calls, not invented |
| 3 | Real photos, case studies and reviews instead of stock | every page has at least one piece of proof |
| 4 | A list of 10–20 customer questions, checked weekly in ChatGPT and Google | a sheet with dates and results |
| 4 | Test speed on a cheap phone | the first screen appears fast even on mobile data |
What not to do
AI panic produces a few common mistakes. Each one costs money and returns very little.
- Generating hundreds of AI articles "for SEO". Google explicitly treats mass-producing pages mainly to manipulate rankings — whether by people or by AI — as a violation of its spam policies (Google Search Central). Ten service pages with real prices and examples beat two hundred rewrites of other people's tips.
- Blocking every bot "just in case". Block Googlebot in robots.txt and your site disappears from both Google's AI answers and regular results. Blocking specific AI crawlers is a deliberate decision with trade-offs, not a default setting.
- Buying a "guaranteed spot in ChatGPT answers". As the Ahrefs study shows, half of citations change by the next day. Nothing here can be guaranteed, so treat promises like "#1 in ChatGPT in a month" with skepticism.
- Hiding prices because "competitors will see". Competitors can get your prices with one phone call. Customers and AI, without prices, simply pick whoever showed theirs.
- Rebuilding the whole site for AI. Usually it's enough to add facts and dedicated service pages to what you already have. A full redesign only makes sense when the problem is the technical foundation.
When edits are enough, and when you need a developer
A business owner can handle much of the above: copy, prices, questions, photos. A developer is needed when the problem sits in the site's foundation.
| Situation | What to do |
|---|---|
| The site runs on Webflow, WordPress or Squarespace, and the text is in the page source | Edit it yourself: copy, prices, new service pages |
| Text and prices appear only in the browser; the page source is nearly empty | Developer: server-side rendering or static generation |
| The form loses inquiries, or spam floods the inbox | Developer: reliable delivery, protection, saving the source of each inquiry |
| You need dozens of similar pages (services × cities, a catalog) | Developer: templates, data, a clean URL structure and internal linking |
| The site is slow on phones even after compressing images | Developer: a speed audit of heavy scripts, fonts and hosting |
I build sites in Webflow when the team wants to edit everything visually, and in Next.js when the site needs custom logic, integrations or sits next to a web app. More on the approach is on the page about website development that brings leads. If you sell products rather than services, the rules differ: catalog data, a Merchant Center feed and purchase tracking. I cover those in how to launch an online store that sells in month one.
What to do this week
Search hasn't disappeared; it has become a conversation. AI now gives part of the answers, and there's no undoing that. But the decision about whom to contact still happens on your website, and that moment is entirely in your hands.
Start with three things: check that your prices and services are visible in the page source, add a "how did you hear about us" field to your form, and start counting inquiries rather than visitors. In a month you'll know which pages sell, and you can invest in them instead of in "traffic" in general.
Was this useful?
– likes · – views
Questions people ask
Does a business still need a website if Google and ChatGPT answer questions themselves?
Yes, but its job changes. AI answers the general questions; the decision about whom to contact still happens on your site, where people check prices, examples, terms and send an inquiry. The website becomes the place where the deal closes, not a reference library.
How do I get my website into AI Overviews?
Google says there are no additional requirements: the page has to be indexed and eligible to be shown with a snippet. In practice the ordinary things matter most: the page ranks in search, states clear answers and facts in its HTML, and its structured data matches the visible text.
Do I need an llms.txt file for AI to cite my site?
Not for Google: its documentation says you don't need new machine-readable files or AI text files. It matters more that the main text of the page is in the HTML your server sends, not painted only in the browser by JavaScript.
How can I see ChatGPT traffic in Google Analytics?
Links in ChatGPT answers usually carry utm_source=chatgpt.com, so GA4 shows them by source. Some clicks arrive with no tag and no referrer and land in Direct, as happened during the October 8–15, 2026 glitch. Add a 'How did you hear about us?' field to your form and count inquiries per page.