How to Optimize Content for Google Gemini and AI Overviews

This article was originally about Google Bard. Bard was renamed Gemini in February 2024. Everything below is updated.

I’ve been testing Gemini against my own pages for a while. I check the logs. I prompt it about my content. Here’s what I keep finding. Gemini is deeply tied to Google’s own index. If Googlebot misreads your page, Gemini probably will too. My working assumption is that fixing how your pages are read improves both.

How to Optimize for Gemini

• Make sure all your content is actually in the HTML. LLM crawlers don’t execute JavaScript. To check what they see, use the URL Inspection tool in GSC. It shows you the HTML that Googlebot received, which is close to what LLM crawlers will retrieve. For a visual check, go to your browser settings, disable JavaScript, and reload the page. Look at what disappears — text, sections, menus. If elements are missing visually, they’re likely missing from the code returned to bots too. Anything important that only appears after JS execution needs to be server-side rendered or in the static HTML.
• Audit what Gemini already knows about you. Go to gemini.google.com and ask it about your brand, your products, your main topics. Is it accurate? Does it say what you’d want a customer to see? Keep in mind that Gemini results are highly personalized, so what you see may differ from what others see. Still worth doing. Inconsistencies in how you’re described across the web — different positioning on LinkedIn, your About page, directories — create confusion that shows up in AI outputs. Clean those up.
• Structure your content around what people actually ask. Think about the real subtopics and questions your audience has. There’s no single formula for headings. Sometimes a question heading works well, sometimes a keyword phrase is fine, and you can always put the question in the text under the heading instead. Many options. What matters is that your content covers the actual subtopics people need answered on this subject, not just the broad keyword.
• Get your metadata and content structure right. How you structure a page affects how LLMs parse it. A well-organized page with clear headings, logical sections, and correct metadata is significantly easier for AI systems to retrieve from correctly. I covered this in depth in my metadata for SEO article and the same principles apply directly to AI visibility. Also make sure your max-snippet meta directive isn’t set too low or to 0. If it is, you’ve told Google not to use your content in snippets, which directly affects AI Overview eligibility.
• Write content that cannot be stolen from you. Generic content gets absorbed into AI training and paraphrased into nothing. What actually gets cited is content where your name is integral to it. Your real tests, your data, your specific observations. Content where removing your name makes it lose its meaning. If your article could have been written by anyone, there’s no reason for Gemini to attribute it to you specifically.
• A recognized author can now be an advantage. Having a named author who is a recognized entity in Google’s Knowledge Graph helps. This doesn’t mean every post needs a byline, but if you or someone on your team is an established entity with a Knowledge Graph presence, associating your content with that author sends a trust signal. It’s not mandatory but it’s worth knowing.
• Keep content updated. Freshness matters for AI Overview citations. Pages that haven’t been touched in years are at a disadvantage regardless of quality. Update key pages when new data is available. Replace vague time references with actual dates.
• Most of what applies to Bing Chat also applies here. Read my Bing Chat optimization article for specifics. In particular, the advice on content chunking and structure carries over directly.
• Your content is yours. Gemini was trained on content people like you produced, often without consent or compensation. You can use robots.txt to block AI training crawlers while still allowing Googlebot. The publisher community is building tools to push back. Make deliberate choices about what you allow to be crawled for training purposes.

How to Find Out What Prompts People Use

There’s no direct way to do this for Gemini. Unlike Bing, where query-level data for AI responses is starting to appear in Bing Webmaster Tools, Google provides nothing equivalent. You cannot see what someone typed into Gemini before it cited your page.

What you can do instead:

• Use GSC as a proxy. People tend to use longer, more complex phrases when querying LLMs than when typing into traditional search. In GSC, filter your Performance report to a short time window like the last 24 hours, then filter for position 1 (or 2) with 1 or more impressions and 0 clicks. Long queries showing up at position 1 with no clicks are a strong signal they were triggered inside AI Overviews. That’s how I find them on my own sites.
• Use keyword research tools filtered for long-tail and question-format queries. These are more likely to trigger AI Overviews than short commercial keywords.
• Use tools like Semrush’s AI Overviews Visibility Checker, Thruuu, or Morningscore. They track which queries trigger AI Overviews and whether your site appears. None of them are perfect but they give you something to work with.

How to Check If the Optimization Worked

The most straightforward measure is whether you’re getting more traffic, or more leads if that’s your goal. AI Overview traffic can show up as direct traffic in Analytics or under other sources. It’s indirect, but if things are working, you should see it somewhere.

Tracking AI outputs directly is costly but possible. According to research, if you send the same prompt 100 times, you can measure how often your brand appears in the answers and get a rough confidence level that you’re being shown for similar queries. Do this for several keywords, not just one. Don’t use the API for this. Query from the actual chat interfaces or use tools that mimic that behavior, since API responses can differ from what users see.

• Impressions without clicks in GSC. If impressions for a page or query are increasing but clicks aren’t following, you may be getting cited in AI Overviews. This is a signal, not proof.
• Branded searches. Search your brand name plus a topic in Gemini and see whether your content is surfaced accurately. Keep in mind results are personalized.
• Third-party trackers. Semrush, Thruuu, Morningscore, and Sitechecker all have AI Overview monitoring features. They query Google on your behalf and track citation patterns over time.

Debugging: What to Do When Nothing Seems to Work

• Start with indexing. Use the URL Inspection tool in GSC to check your pages are indexed and that Google is rendering them correctly. The “HTML that Googlebot received” view is close to what LLM crawlers will see. If your content isn’t there, that’s the problem.
• Check robots.txt and meta directives. Some CDN configurations and security plugins block unlisted bots. Also check max-snippet. If it’s set to 0 or a very low number, you’ve blocked Google from using your content in snippets, which directly affects AI Overview eligibility. More on meta directives here.
• Check if your target queries even trigger AI Overviews. Not all do. Commercial, navigational, and local queries often don’t produce AI Overviews at all. If that’s your category, no optimization will change it.
• Look at what’s being cited instead. For queries where an AI Overview appears but you’re not in it, look at who is. What do those pages have that yours doesn’t? More comprehensive coverage? More recent data? Better structure? That tells you where to focus.
• Check your brand consistency across the web. Inconsistent descriptions across LinkedIn, directories, Wikipedia, and your own site create a blurry entity signal. Tighten it up.
• Wait. Google’s own documentation says recrawling after changes can take days to months. Give it time before concluding something didn’t work.