How to Write Content for AI Overviews: A Guide to Appearing in Google's AI Summaries

How do you write content for AI Overviews?

Writing content for AI Overviews means answering the question in the first paragraph, then backing that answer with evidence, a table and your own experience. Google states that no special format is required; instead, it needs a passage that can stand alone as a citation. Therefore, I write the answer first and the proof second.

I have applied this order on client sites since 2012, long before generative summaries existed. In practice, it started as a featured snippet habit. However, the same discipline now feeds AI Overviews, because the summary engine looks for the same thing: a short, complete, verifiable passage. In this guide, I walk you through that method step by step.

You will also see the numbers behind the shift, the official controls Google offers, and the mistakes I see most often when businesses in Turkey try to "optimize for AI". In short, there is no secret file and no secret markup. There is only clearer writing with a reason to click.

What are AI Overviews and how do they differ from featured snippets?

AI Overviews are generative summaries that Google places above the classic results for some queries. The system combines several sources into one answer and lists those sources as links. A featured snippet, on the other hand, lifts one passage from one page and shows it verbatim.

The difference matters for your writing. A featured snippet rewards a single perfect paragraph; an AI Overview rewards a page that contributes a clear fact to a larger synthesis. Consequently, a page can appear in a summary without holding the first position, as long as it offers a passage the model can use.

• Featured snippet: one source, exact quote, usually one paragraph or list.
• AI Overview: several sources, paraphrased answer, links shown as cards.
• AI Mode: a full conversational interface that Google now runs alongside the classic results.

For example, when someone searches for a definition, the snippet still wins often. Meanwhile, when the question needs comparison or steps, the summary tends to take over. Both draw from the same index, so the same page can serve both.

How does Google choose the sources for AI Overviews?

According to Google's own AI features documentation, AI Overviews use the same index and the same core ranking systems as regular Search. There is no separate crawler, no separate index and no separate application form. If your page is indexable and ranks for related queries, it is eligible.

Semrush's study adds a useful detail here. In its 10 million keyword dataset, 95.32 percent of AI Overviews appeared alongside a "Related searches" block, and 90.03 percent appeared together with "People also ask". In other words, the summary lives in a cluster of related questions, and pages that answer those neighbouring questions get more chances to be cited.

That is why I treat the question list as the raw material. First, I collect the "People also ask" entries for the topic. Then, I check which ones my page answers in a clean passage. Finally, I add the missing ones as their own H2 sections, each with a direct answer underneath.

Which queries trigger AI Overviews today?

AI Overviews began as an informational feature, but that is changing fast. The Semrush AI Overviews study tracked trigger rates across 2025: 6.49 percent of queries in January, a peak of 24.61 percent in July and 15.69 percent in November. Because Google keeps tuning the system, treat any single month as a snapshot.

The intent mix shifted even more. Informational queries made up 91.3 percent of triggering searches in January and only 57.1 percent by October. Meanwhile, commercial queries rose from 8.15 to 18.57 percent, and transactional queries climbed from 1.98 to 13.94 percent.

• Informational: "how to", "what is", comparisons and explanations.
• Commercial: "best", "vs", "review", where buyers compare options.
• Transactional: pricing, packages and "near me" style requests.

For a service business, this means your pricing and comparison pages are now in scope, not only your blog. Therefore, the writing method I describe below applies to service pages as well.

Do AI Overviews really reduce clicks?

Yes, and the effect is measurable. Ahrefs compared 150,000 informational keywords with a summary against 150,000 without one, using December 2025 data. The Ahrefs update found that the presence of an AI Overview correlated with a 58 percent drop in click-through rate for the first organic result: 0.039 without a summary versus 0.016 with one.

Pew Research Center reached a similar conclusion from the user side. In a March 2025 panel of 900 US adults and 68,879 queries, users clicked a result link in 8 percent of searches with a summary and 15 percent without. Clicks on links inside the summary itself reached only 1 percent. In addition, 26 percent of users ended their session right after seeing a summary, compared with 16 percent otherwise.

Both datasets are US and English heavy. In my experience, the direction is the same for Turkish queries, but the ratio can differ, because coverage in Turkish is still narrower. So read these figures as a trend, not as a forecast for your site.

Why is appearing in AI Overviews still worth the effort?

Because the audience is enormous and the alternative is invisibility. At Google I/O, Sundar Pichai reported that AI Overviews now exceed 2.5 billion monthly active users, and that AI Mode passed 1 billion monthly users within a year. A feature at that scale defines how people first meet your brand.

Visibility also compounds in ways a click counter misses. When a summary names your business as a source, the reader sees your brand next to the answer. Even a zero-click session leaves a memory, and that memory shows up later as a branded search, a direct visit or a WhatsApp message.

Example calculation of the trade

Take an example: an informational page with 10,000 monthly impressions in first position. With the Ahrefs values above, a 3.9 percent CTR yields roughly 390 clicks; a 1.6 percent CTR yields about 160. That is a painful gap. However, if the same page adds a comparison table and a tool link, it gains two things: a reason to be cited with a link and a reason to be remembered. These figures are illustrative, not a promise.

How do you write a citable paragraph?

A citable paragraph opens with a definition or a method, carries one idea, runs 40 to 60 words and includes a number or a condition. That is the whole recipe. The summary engine needs a passage it can lift without context; the reader needs a passage that makes sense on its own.

1. Turn the question into an H2, using the words people actually type.
2. State the answer in the first sentence, starting with the definition or the method.
3. Keep the paragraph between 40 and 60 words; use the word counter to check.
4. Add one number, one range or one condition, so the claim is concrete.
5. Place the evidence link right after the paragraph, not three screens later.

Then, test the paragraph by reading it aloud without the heading. If it still answers the question, it is ready. Otherwise, the first sentence is probably an introduction, and introductions never get cited. I run every draft through the readability checker before publishing, because a 40-word sentence is not the same as a 40-word paragraph.

Why do question-shaped headings work so well?

Question headings work because they match the "People also ask" box word for word. Google already knows which questions people attach to a topic; when your H2 mirrors one of them, the passage underneath becomes an obvious candidate. In addition, a question forces you to write an answer instead of a topic summary.

There is a second benefit. A heading such as "Cost of SEO in Istanbul" describes a subject, whereas "How much does SEO cost in Istanbul?" commits you to a figure or a range. Readers notice the difference instantly, and so does the summary engine.

For example, I preview every title and description in the Google SERP preview tool and build the meta tags with the meta tag generator. The heading, the title and the answer paragraph should tell the same story; otherwise, the page looks like three different documents.

How do you avoid the "commodity content" trap?

Commodity content repeats what every other page already says. Google's guide to optimizing for generative AI features highlights this directly: the recommendation is content with a unique, expert or experience-based perspective, rather than a rewrite of common knowledge. A summary engine has no reason to cite the tenth identical explanation.

Here is how I apply it. When I write about a Google Ads setting, I do not only define the setting. Instead, I describe what happened when I changed it on a campaign, what I expected, what actually moved and what I would do differently. That lesson exists nowhere else, so it cannot be a commodity.

Six-question self-check before publishing

1. Could a competitor publish this section unchanged without anyone noticing?
2. Does the section contain at least one observation from my own work?
3. Is there a number, a range or a condition that I measured or verified?
4. Would I say this to a client on a call, in these words?
5. Does the section answer a question that a beginner would ask?
6. Is there a reason to click through after reading the summary?

If you answer "no" to more than two, rewrite the section. Otherwise, publish and move on.

How do you prove experience and expertise for AI Overviews?

You prove E-E-A-T with visible evidence, not with adjectives. Google's quality guidelines use experience, expertise, authoritativeness and trustworthiness as evaluation criteria, and the summary systems inherit the same signals. Consequently, a page with a named author, a real bio and verifiable references has an advantage over an anonymous one.

• Author byline with a link to a real about page that lists years in the field and certifications.
• A references page with clients and results you can defend.
• First-person observations inside the article, clearly marked as experience.
• Contact details, a physical base and a working phone number.
• Dates on the article, including the last review date.

On my own site, every service page and article points back to the same author identity. As a result, Google can connect the article to the person and the person to the field. For a small business, that consistency is cheaper than any backlink campaign and, in my experience, more durable.

Is structured data required for AI Overviews?

No. Google's documentation is explicit that there is no additional requirement and no special optimization for AI Overviews or AI Mode. You do not need a dedicated schema.org type, an "AI text file" or a new machine-readable format. What Google does ask is that any structured data you use matches the visible content on the page.

Structured data still earns its place for another reason. Rich results such as FAQ, article and organization markup improve how the classic listing looks, and that listing still receives most of the clicks. So I keep adding it, but I add it for the classic result, not because a summary demands it.

For instance, I generate article and FAQ markup with the schema generator and then check that every question in the markup appears on the page. Mismatched markup is the fastest way to lose trust with both readers and crawlers.

Should you create an llms.txt file?

You can, but Google says it neither helps nor hurts. The May 2026 generative AI guide names llms.txt and similar files as one of three myths; the official wording is that such files "neither harm nor help" your visibility in Google's AI features. In short, the file is harmless, and it is also not a strategy.

The other two myths deserve equal attention. First, there is no need to chop content into tiny chunks so that "AI understands it better". Second, you do not need to write in a special format only for generative search. Both myths lead to the same bad outcome: pages that read like a database export instead of an explanation.

If a client asks me for an llms.txt file, I add one in ten minutes and then spend the rest of the budget on the answer paragraphs. That allocation reflects where the evidence points.

How do tables, lists and comparisons raise your citation odds?

Tables and lists raise your odds because they compress relationships into a shape the engine can read without guessing. A comparison of two options in prose forces the model to infer structure; a table hands it the structure directly. In addition, the same table gives a human reader a reason to open the page after seeing the summary.

The table below contrasts the old habits I still see in briefs with the citable approach I use today.

Keep the keyword usage natural inside tables too. I run the finished draft through the keyword density tool, because stuffing a table with the same phrase looks mechanical to readers and to Google.

How many topics should one page cover?

One page should cover one main question and its natural sub-questions, each answered independently. Google's guide states that you do not need to split content into fragments for AI; a long, well-organized page remains fine. However, every H2 section must be able to answer its own question without the reader scrolling up.

I use a simple test. Cover the rest of the article with your hand and read one section. If it makes sense alone, it is a self-contained answer. If it depends on "as mentioned above", it is not, and the engine will struggle with it as well.

• One primary question in the title and the first H2.
• Sixteen to twenty-two sub-questions as H2 headings for a long guide.
• No section longer than about 300 words, so each stays liftable.
• Cross-links between sections instead of "see above" references.

As a result, a single guide can earn citations for a dozen related searches, which beats a dozen thin pages competing with each other.

How do you update existing articles for AI Overviews?

Updating an existing article follows an eight-step routine that I apply to client blogs every quarter. It takes one to three hours per article, depending on length, and it usually beats writing something new from scratch, because the page already has history and links.

1. Open Search Console and list the queries where the page appears alongside a summary.
2. Collect the "People also ask" questions for those queries.
3. Rewrite each H2 as a question, using the searcher's own words.
4. Add a 40 to 60 word answer paragraph directly under every H2.
5. Find sections that compare options and convert them into a table.
6. Add an author byline linked to your about page.
7. Refresh dates, statistics and source links; remove anything you cannot verify.
8. Run a readability pass and cut every sentence longer than 25 words.

Then, request indexing again and wait. In my experience, changes appear in summaries within days to a few weeks, depending on how often Google crawls the site. Meanwhile, note the date of the update, so you can compare the before and after later.

Can you remove your content from AI Overviews?

Yes, but only with the same controls that govern classic snippets, and the price is that both disappear together. Google's documentation lists nosnippet, data-nosnippet, max-snippet and noindex as the ways to limit how a page appears in AI Overviews and AI Mode. Setting max-snippet:0 is equivalent to nosnippet.

There is a common misunderstanding to clear up. Blocking the Google-Extended user agent does not remove a page from AI Overviews; that token only affects Gemini training and grounding. Likewise, a robots.txt rule cannot pull a page out of a summary without also blocking it from Search. I build robots files with the robots.txt generator, and I never use it for this purpose.

How do you measure AI Overviews visibility?

You measure it in Search Console first. Google announced in June 2026 a Search Generative AI performance report that shows visibility in generative features such as AI Overviews and AI Mode separately. Google says the report's scope will expand over time, so I treat its early numbers as directional rather than final.

Before that report, and still alongside it, AI Overviews traffic sits inside the "Web" search type in the regular performance report. Therefore, a drop in CTR at stable positions is often the first sign that a summary now appears above you.

• Branded search volume: rising brand queries after a visibility gain.
• Direct traffic: visitors who remember the name and type it later.
• Tagged links in profiles and PDFs, built with the UTM builder, so offline mentions become measurable.
• Assisted conversions in your CRM, where the first touch was a search that produced no click.

In short, count clicks, but also count memory. Otherwise, you will underestimate what the summary is doing for you.

Is anything different for Turkish content and AI Overviews?

The rules are identical, but the playing field is emptier. AI Overviews cover fewer Turkish queries than English ones, and the pool of high-quality Turkish sources is thinner. Consequently, a well-structured Turkish page competes against fewer rivals for each citation, and a local expert has a real advantage over translated content.

Two practical notes follow from that. First, write the Turkish version natively instead of translating an English draft, because the "People also ask" wording differs between languages. Second, mention local context such as Turkish regulations, prices in lira and Istanbul-specific examples; the summary engine tends to favour sources that match the searcher's location and language.

For example, my Turkish articles reference Turkish tax rules and local platforms, while the English versions reference the international sources you see here. The hreflang setup keeps the three versions connected, so each one competes in its own language.

What are the most common mistakes when writing for AI Overviews?

The most common mistake is writing for a machine instead of a reader. I see the same handful of errors in almost every audit, and most of them come from treating the summary as a new algorithm to trick rather than a reader to serve.

• Opening every section with an introduction instead of an answer.
• Splitting one good article into ten fragments because "AI prefers short pages".
• Repeating the focus phrase in every paragraph until the text reads like a list.
• Publishing figures without a source, or with a source that no longer exists.
• Hiding the author, the location and the contact details.
• Blocking Google-Extended and assuming the page is now out of the summaries.
• Ignoring commercial pages, even though their trigger rate is rising fast.

Each of these has a simple fix, and the fix is always the same direction: clearer, more specific, more honest. Above all, resist the urge to create a special "AI version" of your site. Google has said it does not want one.

Does AI-generated content appear in AI Overviews?

It can, because Google evaluates quality rather than the tool that produced the text. The 2023 Google Search guidance on AI content remains in force: content created primarily to manipulate rankings falls under the spam policy, whatever generated it, while helpful content is welcome regardless of origin.

In practice, generated text struggles with the very thing the summary rewards: unique experience. A model can define a term; it cannot tell you what happened on your campaign last March. So I use assistants for outlines and checks, and I write the experience paragraphs myself. That split keeps the article both efficient and citable.

How should your SEO strategy change in the age of AI Overviews?

Your strategy should shift from chasing clicks alone to earning citations, clicks and memory together. The pages that win now answer first, prove second and leave a reason to visit third. That order is the entire method in this guide, and it works for a blog post, a service page or a pricing table.

Two adjustments matter most. First, plan content around question clusters rather than single keywords; I build those plans together with clients through SEO consulting. Second, balance any informational traffic loss with paid demand capture, which is where careful Google Ads management earns its keep.

If you want a second pair of eyes on your current articles, get in touch. I work directly, without intermediaries, and the first conversation is about your questions, not mine.