Service pages for AI search: eight fields every page needs

A service page describes exactly one service. If you offer several, each one gets its own page once it’s genuinely different for buyers; mere variations belong on a single page as sections. A pricing page uses the same fields but groups several services. Delivered HTML is the page code the server sends before a browser runs any scripts.

This how-to isn’t about your website as a whole. How to get your business recommended by ChatGPT walks you through access for AI crawlers (the programs AI providers use to fetch pages), consistent company facts and mentions on other websites. Our guide to GEO lays out what’s actually proven about generative engine optimization (GEO), the work of improving AI visibility.

The title is the page name shown in search results, and the H1 is the page’s main heading. By path to a price, we mean how a buyer finds out what the service costs, with or without a published figure. We put the fields together from Google’s and Microsoft’s guidance and from the questions buyers ask before they get in touch. No study has tested the set as a whole. The step Do your pages answer your customers’ questions up front? in our ChatGPT how-to shows how to write each field as a direct answer.

What you don’t need: For Google Search, you don’t need special files such as an llms.txt or special markup, and you don’t have to break your text into small chunks. Structured data, meaning machine-readable details in the page code, isn’t required for Google’s AI features either. All of this comes from Google’s guide to generative AI search, as of Jul 10, 2026 (external site). As Microsoft describes it, Microsoft Copilot breaks content into smaller pieces when processing it, and the company recommends clear headings and lists (Microsoft Advertising, Oct 8, 2025 (external site)). The two fit together: a clearly structured page is enough, and you don’t need a special format for AI.

None of this comes with a guarantee. Google says that even pages that meet all its requirements aren’t guaranteed to be shown, and Microsoft says that nothing guarantees selection for an AI answer.

Done when the eight fields for one of your services are written down as notes on a single sheet. You’ll fill them in over the next steps.

Why: According to Microsoft, AI systems use the title, description and H1 to interpret what a page is about. You don’t have to cover every wording, because Google says AI systems also understand synonyms. Across 1.4 million ChatGPT prompts, Ahrefs measured how closely page titles match the prompt. Titles of cited pages scored 0.602 on a similarity measure, and titles of pages that weren’t cited scored 0.484. A higher score means the title is closer to the prompt. That’s a correlation, measured in a single system by a vendor that sells its own AI visibility tools. The scores only approximate how ChatGPT works, and they don’t prove cause.

Source: Ahrefs (external site), Apr 15, 2026 · vendor study · 1.4 million prompts, ChatGPT only

How to do it:

1. Collect the words customers use for the service from emails, requests and intro calls.
2. Put the most common term in the title, the H1 and the first sentence. Related terms belong in the text.
3. Read the first sentence on its own. It should explain the service without the rest of the page. Microsoft recommends self-contained sentences like this, too.

Example: On our customer portal page, the overview opens with this sentence: “Datafalke, based in Bottrop, Germany, builds customer portals for B2B companies.” It names the provider, the location, the service and the audience in one sentence.

Common mistake: You create a separate page for every wording. Google treats that as a violation of its spam policies (external site) if the main purpose is to influence rankings or AI answers.

Done when the title, the H1, the first sentence and, if you have it, the name in your structured data all name the same service.

Region: Spell out your location and service area, for example “Installation in North Rhine-Westphalia, planning by video nationwide.” In structured data, the area is called areaServed. Schema.org’s definition (external site) says it’s the geographic area where a service is provided. The markup adds to the sentence in your text and doesn’t replace it. For on-site services, Google also names Google Business Profile as a way to become visible in AI answers. The step Are your company facts clear and consistent everywhere? in our ChatGPT how-to shows how to keep your details the same everywhere.

Audience: Describe your customers with traits readers can check against themselves, such as company size, industry or a specific situation. Also say who the service isn’t right for.

Example: Under “Good fit,” our GEO services page includes “Companies with a website that AI answers leave out.” Under “Not a fit,” it includes “Anyone expecting guaranteed mentions.”

Common mistake: Each city gets a page that only funnels visitors to one central landing page. Google lists such pages as an example of doorway abuse (external site), meaning pages built only to pass visitors along.

Done when the region and audience appear on the page in full sentences, and one sentence says who the service isn’t right for.

Why: According to Microsoft, lists, numbered steps and tables break even complex details into reusable segments. Words like “innovative,” on the other hand, mean little without a measurable fact behind them. Your readers benefit the same way: they can see what’s included without having to ask.

How to do it:

1. Write each part of the service as its own list item.
2. Add a quantity, a result or a limit to each item.
3. Say what isn’t included.

Example: Under “Monthly measurement,” our GEO services page lists “25 questions across four AI systems, three runs each, and a report that includes drops.”

Common mistake: The scope sits in a PDF, an expandable menu or an image. Microsoft advises against this: AI systems may not render hidden content, PDFs often lack structural signals such as headings, and text is recognized less reliably from images.

Done when every list item names something checkable and the whole list is in the page text, with no click needed.

What the data shows: AirOps analyzed 217,508 pages that ChatGPT retrieved for 7,500 commercial prompts. For prompts seeking to confirm a single detail such as price, pages with at least 15 price points had an 11.0% higher likelihood of being cited. That applies only to ChatGPT and only to pages ChatGPT had already retrieved. The study’s examples come from the software market, and AirOps itself sells tools for this kind of content. It’s a correlation. It doesn’t prove cause, and it says little about service businesses in Germany.

Source: AirOps (external site), Apr 1, 2026 · vendor study · 7,500 prompts, ChatGPT only, correlations

How to do it:

1. Choose the path to a price that fits your service and write it as a full sentence next to the scope.
2. Say what drives the price and what the next step costs, for example a free intro call.
3. Only mark up a price in your structured data if it’s visible on the page. The properties of Offer on Schema.org (external site) include price and priceCurrency for this. Google requires marked-up content to be visible to readers (external site), and its overview of supported structured data (external site) lists no dedicated feature for the type Service. Our article Structured data covers the basics.

For typical market ranges for GEO, see our cost guide How much does generative engine optimization (GEO) cost? If you also sell to consumers, have a lawyer review how you present your prices. Our guide to publishing prices shows how to write a price statement as a full sentence with an as-of date and keep it current.

Example: Our service pages don’t currently give our own prices. Instead they say “Fixed price after analysis,” and our GEO services page says we quote the fixed price “after a short analysis.” Of the 15 pages, 14 list what drives the price. So we can’t show on our own pages whether published figures would change anything about mentions in AI answers.

Common mistake: “Price on request” with nothing else. Readers then learn neither how to get a price nor what drives it.

Done when one sentence on the page says how prospects get to a price and what drives it, and the markup says nothing the text doesn’t.

How to do it:

1. Write down the steps in the order they actually happen.
2. For each step, say what exists afterward, such as a quote, a draft or a report.
3. Give durations as estimates and add that the schedule in the quote is the one that applies.

Example: Our service pages list the steps under headings such as “How the project runs” or “How we work together.” Every list starts with the intro call. Each step names a result, and most also give a duration.

Common mistake: The process appears only in a graphic. Microsoft advises against putting key information only in images, because text in images is recognized less reliably.

Done when every step has a name and says what exists afterward.

Why: Google advises creating content from your own knowledge and experience rather than repeating what others have already said. Microsoft writes that a claim like “next-gen” without context leaves AI systems unsure how to classify it. Both points favor details a reader can verify.

How to do it:

1. Name approved references with a time period and a result.
2. Show work samples with permission, such as an anonymized report.
3. State commitments that can be checked, such as written acceptance criteria or the documents customers receive at the end.
4. Link to your expert evidence: sources, standards with their numbers or the profile of the person responsible.

Example: None of our 15 service pages lists a customer reference yet, because we only name customers with their written permission. Instead, all 15 name written acceptance criteria and the documentation customers get.

Common mistake: Rows of logos, star ratings or percentages with no source and no time period. No one can verify them.

Done when every key claim on the page has evidence a reader can follow.

How to do it:

1. Name the person who actually answers, with their name and role.
2. Build the contact route as a normal link or an HTML form.
3. Say what happens after someone gets in touch. Don’t promise response times you can’t keep.

Example: Our service pages have a “Your contact” section with the owner’s name. Requests go through a plain HTML form, and the email address is a link in the page footer.

Common mistake: The contact route only appears through a script, such as a chat window or an embedded booking widget. Our ChatGPT how-to explains why that matters for AI crawlers in the step Are your services visible in the HTML, without JavaScript?

Done when the name and at least one contact route are in the page source.

Without a terminal, open the page source. In most browsers on Windows, that’s Ctrl+U. In Chrome on a Mac, it’s Command+Option+U. Then search it with Ctrl+F (Command+F on a Mac) for your service name, your region, your path to a price and your contact details. Our article JavaScript and AI crawlers explains how delivered HTML differs from the fully rendered page.

The terminal is faster: on macOS and Linux it works directly, and on Windows it works in Git Bash. The curl command downloads only the HTML and doesn’t run JavaScript. Replace the address and the words in step 4 with your own:

URL='https://www.example.com/services/example/'

# 1. Status and final address
curl -sL -o page.html -w '%{http_code} %{url_effective}\n' "$URL"

# 2. Service name in the title and H1
grep -oE '<title>[^<]*|<h1[^>]*>[^<]*' page.html | sed -E 's/<[^>]+>//'

# 3. H2 headings: audience, process, pricing, contact
grep -oE '<h2[^>]*>[^<]*' page.html | sed -E 's/<[^>]+>//'

# 4. Count words in the HTML: use your own words
for word in 'fixed price' 'Ruhr area' 'sign-off'; do
echo "$word: $(grep -oi "$word" page.html | wc -l)"
done

# 5. Structured data: Service, and Offer only with a visible price
grep -oE '"@type" *: *"(Service|Offer)"' page.html | sort | uniq -c

# 6. Contact routes: email, phone, form target
grep -oiE 'href="(mailto|tel):[^"]+"| action="[^"]*"' page.html | sort -u

How to read the output: Step 1 should show 200 and the address you expect; otherwise the server redirects or the page is missing. Steps 2 and 3 show whether the service name and the fields appear as headings. If a heading contains another tag, the output stops there. Step 3 lists every H2 on the page, including those in the navigation and footer; it doesn’t show subheadings (H3). Step 4 counts your words. Step 5 shows which types the JSON-LD names. JSON-LD is the block that holds the structured data. Step 6 lists email and phone links and the target of a form.

Common mistake: The search doesn’t find a word even though it’s on the page. A soft hyphen in the source (­ or the invisible character U+00AD) splits the word. That happens on our German business software page, where the H1 contains “Verwaltungssoftware.” Conversely, grep also counts matches inside the JSON-LD. So a match only means the word occurs somewhere in the HTML. To see whether it’s visible, use Ctrl+F on the displayed page.

Done when the commands show all eight fields for your most important service page, or you know which one is missing. Repeat the check after every major change to the page.