Himanshu SharmaSoftware engineer
Let’s talk

AI Search: AEO and GEO · Lesson 20 of 20

Project: make an FAQ page AI-ready

A hands-on project taking a small FAQ page through answers, structure, markup, crawler access and honest measurement, using this site.

20 minMeasure, behave, buildUpdated 14 November 2026

Awaiting technical review. This lesson is readable but not yet listed in search engines.

By the end of this lessonProduce a small FAQ or answer page that is crawlable, clearly written, consistently marked up and measured honestly, without any promise of citations.

Putting it together

After this project you will have taken one small page from "fine" to "clear, accessible to crawlers, consistently described and measurable", and you will know exactly which parts were evidence-based and which were hypothesis.

You can do the project on a page of your own. The worked example below uses a page that really exists in this repository: the URL Encoder & Decoder tool page at /tools/url-encoder/. Its FAQ items come from the site's tool catalogue. I describe only what the code shows; I make no claim about how the page performs.

Step 0: set the rules

Write them at the top of your notes:

  • No promise of rankings, citations or traffic.
  • Every change gets a reason: reader benefit, vendor document (named) or "hypothesis, low cost".
  • Nothing invented: facts, quotes, reviews, credentials.

Step 1: eligibility (documented)

Check the page can be used at all.

  1. Does it load with a normal success status?
  2. Is it free of noindex and nosnippet? (Google's AI features page says a page must be indexed and snippet-eligible.)
  3. Is the canonical link the page's own preferred address?
  4. Is it listed in the sitemap and linked from other pages?
  5. Is the content in the initial HTML (view-source)?

For the example: the tool page is built as static HTML, the site's seo.ts sets canonical, description and social tags, and scripts/seo-audit.mjs checks title, description, canonical, one h1 and valid structured data on every exported page. Whether the live deployment passes is for Himanshu to confirm with the audit; I have not run it.

Step 2: write the answers (established practice)

The example FAQ begins with the question "When should I encode a component instead of a full URL?" and answers with the rule first: encode a component for a query value or path segment, and encode a full URL only when you want the whole address made safe while keeping its structure. That is the direct-answer pattern from lesson Direct answers: question as heading, plain answer first.

Do the same for your page:

  1. List 5 genuine questions from real users.
  2. Write each answer so it makes sense alone.
  3. Add one caveat where truth needs it.
  4. Add one piece of original value, such as a worked example from your own use. For the URL tool, the tool's own sample (the input a b&c=d becomes a%20b%26c%3Dd) is a real, checkable example that you can run yourself in the URL encoder.

Step 3: structure (established practice)

Use one h1, h2 headings for question groups, real lists for steps, a table only if it compares something. Keep the FAQ in the HTML. A native details element works.

html
<main>
  <h1>URL encoding: questions and answers</h1>
  <h2>When should I encode a component instead of a full URL?</h2>
  <p>Encode a component for a query value or path segment. Encode a full URL only when you want the whole address made safe while keeping its structure.</p>
  <h2>What does %20 mean?</h2>
  <p>It stands for a space. An ampersand inside a value is written %26.</p>
</main>

Step 4: markup that matches (documented, with a caveat)

Add FAQPage JSON-LD only if it matches the visible text exactly, and know its limits: Google's documentation states that FAQ rich results no longer appear in Search as of May 2026, and no vendor documents that this markup causes citations. It is optional. If you add it, generate it from the same data as the page, which is what this site does for tool pages.

json
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "When should I encode a component instead of a full URL?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Encode a component for a query value or path segment. Encode a full URL only when you want the whole address made safe while keeping its structure."
      }
    },
    {
      "@type": "Question",
      "name": "What does %20 mean?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "It stands for a space. An ampersand inside a value is written %26."
      }
    }
  ]
}

The first question is the one on the real page; the second is an added teaching example, not on the site. Paste it into the JSON formatter to prove it parses, then compare each string to the page.

Step 5: access (documented)

  1. Open robots.txt and confirm the page is not disallowed. This site's file disallows /admin/ and /api/ only.
  2. Decide on training opt-outs separately and deliberately (lesson robots.txt rules for search and AI crawlers). The example site names no AI crawlers; that is a pending decision, not an accident to fix silently.
  3. Check no firewall blocks crawlers.
  4. Skip llms.txt unless you have your own reason: Google says it is unnecessary for its features.

Step 6: trust (established practice)

Add a byline and a date that reflect real changes, a link to who runs the site, and a contact route. Do not add credentials you cannot show.

Step 7: measure honestly

  1. Verify the site in Search Console and Bing Webmaster Tools if you have not.
  2. Note the date of your changes.
  3. Run a 10-prompt manual test before and a month after, with the recording template from lesson Measuring AI visibility honestly.
  4. Report counts, accuracy and limits, never promises.

Final checklist

Check Evidence type
Indexable, crawlable, canonical correct Documented
Answer first, question headings Established practice
Content in the initial HTML Established practice, documented advice from Google
FAQ markup matches text Documented rule; benefit unproven
No hidden text or instructions Documented spam policy and ethics
Prompt test recorded Your own data, small sample

Common mistakes

  • Starting with markup instead of the answers.
  • Adding questions that nobody asked.
  • Marking up an answer that differs from the visible one.
  • Declaring success from one assistant answer.
  • Treating the checklist as a guarantee.

Exercise

  1. Choose a page of your own (or the URL encoder page), and complete steps 1 to 4 in writing.
  2. Run the before test of 10 prompts and store the results.
  3. Write the one-page report: what you changed, what evidence label each change has, what you measured, and what you do not know.
Show a sample answer

A good report ends with open questions, for example: "We do not know whether any assistant uses this page as a source. Next review: in one month, repeating the same prompts." A report that ends with a promise has missed the point of the course.

Previous: Risks, ethics, hallucinations and opting out. Back to the course start.

Key points

  • The project applies the whole course to one small page: answers, structure, markup, access, measurement.
  • The real example is the FAQ on this site's URL Encoder tool page, a page that already exists in the repository.
  • Every change is justified by a reader benefit or a vendor document, not by a promise.
  • Markup must match the visible text; FAQ markup earns no Google rich result as of 2026-11.
  • Success is a checklist you can verify, not a citation you cannot control.

Check yourself

1. What is the first step in the project?
2. How should the FAQPage JSON-LD relate to the visible FAQ?
3. Which result can the project honestly promise?
4. After publishing, what is a sound measurement step?

Free tools for this lesson

Final exam and certificate

Finish all 20 lessons to unlock a 15-question exam on AI Search: AEO and GEO. Pass it and you get a certificate with a public link you can share.

Sign in to take the exam