How to Build a Page That Gets Cited in AI Summaries

Summary: AI summaries take the top of the results page, so the click-through rate can fall even while impressions grow. The pages that end up inside the answer are usually the ones that answer the question directly, are marked up with structured data and pass the technical thresholds.

What changed on the results page

A searcher used to look at ten blue links. Now a ready answer sits at the top of the screen, and the sources appear as short citations. That has two consequences. First, your content can be shown inside the answer, which is a new place to be seen. Second, if the question is settled right there at the top, the user can leave without touching a single link.

What we measured on our own site

It is better to show this with our own data than with guesses. According to Search Console, saytpro.az collected 21,269 impressions and 358 clicks in the last 90 days. In the 90 days before that it had 14,493 impressions and 311 clicks.

• Impressions grew 47 percent
• Clicks grew 15 percent
• The click-through rate fell from 2.15 percent to 1.68 percent

So traffic did not drop; it grew. But the share we take from each impression got smaller. The cause can't be pinned on a single factor, since the mix of queries has also widened. Still, the direction is clear: being pleased with the number of impressions is not enough, you have to measure what each impression earns.

How generative engines choose content

When it builds an answer, the model does not read the whole page; it takes the piece it needs. That is why the structure of the page matters more than polished wording.

• The heading is the question, the first paragraph is the answer. Let the heading repeat a real question and the answer come right after it. An answer that waits three paragraphs after the introduction is usually not taken.
• One paragraph, one idea. When two ideas share a paragraph, the piece loses its context once it is lifted out and becomes useless.
• Checkable specifics. Numbers, dates, names, prices. "Very fast" can't be quoted; "under 2.5 seconds" can.
• Your own data. General information repeated everywhere does not make you the one to pick. A number you measured yourself and an example from your own work exist nowhere else.

Structured data, the layer machines read

Structured data turns the information on a page into a format a machine can read without ambiguity. The company name, contact details, an article's author and date, a service's price: instead of getting lost in the text, each of them is marked up on its own.

Where it helps most: company and contact details, the author and date on article pages, service and product pages, and the breadcrumb trail. One condition: the markup must match the content visible on the page. Marking up something that is not on the page brings nothing.

Trust signals, or E-E-A-T

The model wants to know who wrote the page. This is not a question of design; it is a question of evidence.

• Articles with a named author and an update date
• A company page with a real address, phone number and team
• Case studies: what was done, how long it took, what came of it
• Mentions of the company name on other sites
• Prices and terms stated openly

The last point is the one skipped most often. A page that hides its prices never ends up inside an answer, because it offers nothing concrete to answer with. That is exactly why our own prices are public: a landing page from 800 ₼, a corporate website from 1,300 ₼, an online store from 1,800 ₼, and monthly SEO at 700 ₼.

Technical thresholds

Here you should work with the thresholds Google itself has published, not with an estimated score:

• Loading of the largest content element, LCP, under 2.5 seconds
• Response to a user's tap, INP, under 200 milliseconds
• Shifting of page elements, CLS, under 0.1

These metrics are measured on mobile devices, on real user data. Scoring well on desktop and failing on mobile is the most common case.

Checklist

1. On your five most important pages, are the headings real questions or ad slogans?
2. Is the answer in the first paragraph under each heading?
3. Do the company, article and service pages have structured data?
4. Are the author and the update date visible?
5. Do LCP, INP and CLS pass on mobile?
6. Does the price page give concrete numbers?

SaytPro is a team of developers working in Baku since 2023, and our own team does the work. Monthly SEO costs 700 ₼, and after handover you get 2 months of warranty and technical support. Write to us for a free consultation about where your site stands today.