7 technical SEO areas that matter for AI search in 2026
Google is once again saying that AI search is built on classic SEO (search engine optimization), not on tricks "for chatbots." We gathered seven technical areas and compared what is known about their impact on visibility with what is just hype.
A foundation instead of tactics with an expiry date
In a conversation with SEO consultant David Iwanow, Google's John Mueller addressed the trend of "listicles." These are rankings like "top 10 companies in the industry," which companies publish so that AI chatbots point to them as the best choice. According to Mueller, this is an old pattern. One tactic starts working, everyone copies it, half of them do it badly, and eventually the systems learn to treat it as spam.
So if you go all out and say, my whole strategy is listicles, then at some point that's going to switch off and you're left with nothing really.
Mueller does not consider every list to be spam, but a list has to add value. In his view, AI systems "build on SEO," so technical SEO will remain very important. AI crawlers fetch pages a little differently, because some collect data for model training while others look for the freshest content. Mueller treats this as a variation of familiar technical SEO, not its end. That was the assumption we started from when choosing the areas for this guide. We assessed each of them using four criteria:
- whether Google's position is directly confirmed or based only on industry data,
- what impact the area has on visibility in AI Overviews (AI-generated summaries above Google search results) and AI Mode (the conversational AI mode in Google Search),
- how much implementation costs (time, tools, developer work),
- whether a small business owner can do it alone or needs a team or an agency.
One caveat. Some of the data about AI crawlers, files for AI, and structured data comes from blogs run by companies selling SEO tools. We label such data as industry data because no one has independently verified it.
1. Crawler access and indexing
Who it's for: every website, from a one-person store to a major portal.
What it does: AI answers use the same foundation as classic search. A page that a crawler cannot fetch or index will not appear in any of these places.
Limitations: allowing crawlers in the robots.txt file (instructions for search engine crawlers) does not guarantee that AI will cite the page. Allowing crawlers that collect data to train models is a separate decision, including whom you are giving your content to.
Cost: the indexing report in Google Search Console is free, and a dozen or so minutes per month is enough for it. Server log analysis, meaning checking which crawlers actually visit the site, usually already requires a specialist.
2. JavaScript rendering
Who it's for: websites built as SPAs (single-page applications that load content with scripts in the browser) on React, Vue, or Angular. This also applies to online stores using such solutions.
What it does: according to industry analyses, GPTBot, OAI-SearchBot, ClaudeBot, and PerplexityBot fetch only static HTML and do not execute JavaScript. Googlebot renders pages in the Chrome browser, so within Google itself the problem is smaller. A page that shows content only after scripts are run may be empty for some AI systems. Industry blogs cite data from Vercel and MERJ, according to which 69 percent of AI crawlers do not execute JavaScript. We did not reach the original source of this number.
Limitations: unblocking crawlers in robots.txt will not change anything here. A change in the site's architecture is needed, meaning SSR (server-side rendering) or pre-rendering, for example in Next.js or Nuxt frameworks.
Cost: you can do a simple test yourself. Just view the page source or disable JavaScript in the browser and see whether the content remains. Rebuilding a site for SSR is already a project for a development team.
3. Page speed after the Lighthouse 13.6.0 update
Who it's for: everyone, but especially teams and agencies that report results to clients or management.
What it does: PageSpeed Insights already uses Lighthouse 13.6.0, released on October 27, which simulates page loading differently. Barry Pollard from the Chrome team wrote that simulated FCP (the moment the first content appears) and LCP (the moment the largest element is rendered) are now more accurate and usually come out better. The change will more often affect sites based on large amounts of JavaScript. One developer pointed to a clear improvement in results for SPA applications, giving react.dev as an example, and Pollard confirmed it. Some of the fixes were initiated by developer Ivan Akulov. On GitHub, he showed a test page built in Framer for which the old simulation overstated LCP.
Limitations: the score in PageSpeed Insights comes from a simulation on one device and under fixed network conditions. According to Google's documentation, a good lab score does not mean that real users also have a good experience. Actual visits from the last 28 days are shown by CrUX (Chrome User Experience Report) data. On Lighthouse GitHub, there is still an open issue about cases where simulated LCP greatly exceeds measured LCP, mainly on large JavaScript-heavy sites.
Cost: PageSpeed Insights is free. The Lighthouse team expects the new version to reach Chrome 159 developer tools, whose stable release is scheduled in the Chrome calendar for December 8. Proper speed optimization is developer work.
4. Structured data
Who it's for: stores, local businesses, publishers—in other words, anyone who has organized information about products, prices, addresses, or articles.
What it does: brings order to data and enables rich results in classic search.
Limitations: Google does not require any special schema.org markup for a page to appear in AI Overviews or AI Mode. Industry blogs describe an Ahrefs study on 1,885 pages in which adding markup did not produce a statistically significant change in citations by AI. The results were +2.4 percent in AI Mode, +2.2 percent in ChatGPT, and -4.6 percent in AI Overviews. We have not seen the original report. According to industry reports, Google is also phasing out some structured data types, although existing markup does not need to be removed.
Cost: in popular content management systems (CMS), a plugin adds basic markup, so a small business can handle it on its own. Large product catalogs require developer work.
5. The llms.txt file
Who it's for: currently no one who is counting on visibility in Google.
What it does: according to industry reports, in June 2026 Google clarified in Search Central documentation that the search engine ignores llms.txt files. They affect neither ranking nor AI Overviews. Mueller reportedly compared this file to the old keywords meta tag, which search engines have ignored for more than a decade because a site owner can put anything in it.
Limitations: industry blogs cite an SE Ranking study of 300,000 domains. According to it, about 10 percent of domains have llms.txt, and 97 percent of those files have never been fetched by any crawler. We were unable to access the report itself.
Cost: the file itself takes 15 minutes to create. The real cost is the agency that sells it as "AI optimization." Mueller advises small businesses to keep asking providers what they actually do. He compares today's questionable tactics to the old practice of keyword stuffing and white text on a white background.
6. Agentic Resource Discovery (ARD)
Who it's for: companies that provide APIs (interfaces for other programs), tools, or services that AI agents (programs that independently perform tasks on behalf of the user) can use. A regular store or blog currently has nothing to describe this way.
What it does: ARD is an open specification that Google published on June 17. It is supported by Microsoft, GitHub, Nvidia, Salesforce, and Cisco, among others. It involves placing a manifest in a fixed location on the domain, currently /.well-known/ard.json. Lighthouse 13.6 looks for it in several places: in the Agentmap line in robots.txt, in a link tag with a rel attribute value of ard, in the Link header, and in the ard.json path. Only then does it check the older ai-catalog names.
Limitations: the standard is very young and has already changed the file name within a few months. According to an analysis cited by an industry blog, fewer than 3,000 domains had a correct file 60 days after launch.
Cost: the JSON file itself is simple. A sensible description of services for agents, however, requires a technical team and a well-thought-out API offering.
7. Measuring AI traffic in Search Console
Who it's for: anyone who needs to show whether presence in AI Overviews translates into visits.
What it does: on August 31, Google announced that the Generative AI report in Search Console was being rolled out to all websites. It shows total link impressions in AI Overviews and AI Mode, broken down by pages, countries, devices, and dates. It does not show clicks, CTR (click-through rate), or queries. Clicks from AI features go into the regular performance report and cannot be separated out.
Limitations: Google's Gary Illyes wrote that the team is "super interested" in providing more data, but has nothing to announce. The UK regulator CMA (Competition and Markets Authority) requires Google to provide publishers with impressions, clicks, and CTR from AI features by December 3, for example through Search Console. The rules apply to results shown to users in the UK. It is not known whether Google will make this data available in other markets as well.
Cost: the report is free. For now, reporting on AI visibility has to be based on impressions, without promising anyone a number of clicks.
What to choose if you run a small business, and what in a large organization
If you run a small store or work as a freelancer, start with what you can do yourself at no cost. Check the indexing report and the Generative AI report in Search Console. Test the site without JavaScript. Add basic structured data with a plugin. You can ignore llms.txt and ARD for now, and treat "chatbot SEO" services with great caution.
If you are responsible for a large site, especially one built as an SPA, the priority is server-side rendering and log analysis for AI crawlers. Organize speed reports by Lighthouse version and base them on data from real users. ARD makes sense when the company provides APIs or services for agents. It is worth planning AI traffic measurement so that you can use the data right away when Google starts showing it.