How to Add FAQ Schema for AI Overviews the Right Way
Your FAQ enhancement report vanished from Search Console, and now you are wondering whether the markup was ever worth the effort. Google retired FAQ rich results on 7 May 2026, and the SEO industry split into two loud camps inside a day. This guide covers how to add FAQ schema for AI Overviews, what the evidence says it really does, and whether your existing markup should stay. You get a working template, install steps for four platforms, and a validation routine that survives the June 2026 tool changes. To add FAQ schema for AI Overviews, write real questions with short visible answers, wrap them in FAQPage JSON-LD, paste the script into your page head, then validate with the Schema Markup Validator. Google retired FAQ rich results in May 2026, so the visible question-and-answer content now does the heavy lifting. One thing needs settling before any code. FAQ schema and FAQ rich results are two different things, and almost all the May panic came from treating them as one. Affordable SEO Services put this walkthrough together after that documentation change, and the order below is deliberately the order you should work in. What Is FAQ Schema and Does It Still Work in 2026? FAQ schema is a block of code that labels which parts of your page are questions and which parts are the answers. It uses the FAQPage type from Schema.org, and each pair sits inside a Question object with an acceptedAnswer attached to it. Nothing about it changes what a visitor sees. The markup is a translation layer for machines, not a design feature, and that distinction explains most of what follows. You could write the same semantic markup as microdata, using attributes sprinkled through your HTML, but Google recommends JSON-LD and every tool in this guide assumes it. Microdata is harder to maintain and easier to break during a template change. Does it still work? That depends on what you mean by work. As a Schema.org type, it is completely valid, and Google still documents it. As a way to win rich snippets and extra space in the results, it is finished. What survives is the underlying job. Question and answer pairs are the most extractable content shape on the web, which is why featured snippets, voice search, and AI Overviews all reward the same formatting. What Changed When Google Dropped FAQ Rich Results? Google added a deprecation notice to the top of its FAQ structured data documentation on 7 May 2026, and FAQ rich results stopped appearing in Search. There was no blog post and no webmaster announcement, just a small label on a developer page. Within a day, one half of the industry declared FAQ schema dead, and the other half declared it more important than ever. Both readings were wrong, because each one confuses the feature with the format. Which three dates matter? If any dashboard or Looker Studio connector pulls FAQ appearance data through that API, the pull is about to break. Archive the historical numbers this week. Why did Google pull the feature? Google gave no public reason, but its own three-year pattern tells the story. In August 2023, the company restricted FAQ rich results to well-known, authoritative government and health sites, which removed eligibility for almost every commercial page. HowTo rich results went from desktop the following month. Seven more structured data types were retired in June 2025, including Course Info, Claim Review, and Estimated Salary, on the reasoning that they were not widely used and no longer added much for users.The thread running through all of it is abuse. Once a rich result gets bolted onto every page by a plugin, it stops describing the page and starts decorating the listing. Should you delete your existing FAQ markup? No. Google has said unused structured data does not cause problems for Search, and the May documentation update specifically notes the markup can be left in place. There is no penalty for keeping valid FAQPage code where it is.What it does cost you is maintenance. If your markup no longer matches the visible page, that is a reason to fix it or pull it, and it always was. Does FAQ Schema Actually Get You Cited in AI Overviews? Probably not on its own. The strongest causal test available found no meaningful citation uplift from adding schema at all, which is the opposite of what most guides on this topic will tell you.Start with the number everyone quotes. Ahrefs analysed 6 million URLs and found that pages cited by AI were roughly three times more likely to carry JSON-LD. Frase, citing an SGA Index analysis, puts FAQ schema pages at 3.2 times the citation rate of pages with no structured data. Correlation was never really the question, though. Sites that bother with schema also publish better content, earn more links, and maintain their templates properly, so the markup may simply be a marker for everything else they do well. Ahrefs went and ran the harder test. Louise Linehan and Xibeijia Guan tracked 1,885 pages that added JSON-LD for the first time between August 2025 and March 2026, matched them against 4,000 control pages on other domains with similar prior citation levels, and ran a difference in differences analysis across Google AI Overviews, AI Mode, and ChatGPT. Google AI Mode moved 2.4 percent. ChatGPT moved 2.2 percent. Both sit inside the noise. AI Overviews came in at minus 4.6 percent, small but statistically real. A nick, not a crater, and in the wrong direction. Their own summary of it was blunt. Adding schema produced no major uplift in citations on any platform. A separate experiment from searchVIU asked a narrower question. It tested whether five major AI systems read schema when fetching a page live, and ChatGPT, Claude, Perplexity, Gemini and Google AI Mode all pulled only the visible HTML. JSON-LD, hidden Microdata and hidden RDFa were ignored at the point of retrieval. Mark Williams-Cook demonstrated the same thing from another angle. He published
What Is llms.txt and Does Your Site Need One in 2026?
llms.txt is a plain text file, written in Markdown, that sits at your site’s root and points AI tools to your most important pages. Jeremy Howard of Answer.AI proposed it in 2024. No major AI provider has confirmed that it reads the file, so most sites do not need one yet. Ask ChatGPT about your own business, and you may not like the answer. It might name the wrong services, quote pricing you retired last year, or skip your best pages completely. That gap is why llms.txt keeps coming up in technical SEO conversations. Here is what the file does, what the data says, and whether your site needs one. What is an llms.txt file? An llms.txt file is a curated list of your best content, written for machines rather than people. You save it as Markdown, name it exactly as loads at yourdomain.com/llms.txt, and host it at your root. The reasoning behind it holds up well. Large language models do not index your whole site the way Google does. They pull pages in real time, work inside limited context windows, and struggle with anything buried under heavy JavaScript. The official llms.txt specification argues that models benefit from concise, expert-level information gathered in one accessible location, and that converting cluttered HTML into clean text is both difficult and imprecise. # Site title> A brief description of the documentation site.## Docs– [API](https://example.com/docs/api.md): Endpoint list and usage- [Install](https://example.com/docs/install.md): Setup steps- [Getting started](https://example.com/docs/start.md): Intro guide## OpenAPI Specs– [openapi](https://example.com/docs/openapi.json)## AsyncAPI Specs– [asyncapi](https://example.com/docs/asyncapi.yaml) So the file is meant as a shortcut. Rather than hoping a model stumbles onto your pricing page, you hand it a map of what matters. What does an llms.txt file actually look like? Here is a short llms.txt example following the structure the specification describes. An H1 for the site name, a blockquote summary, then H2 sections holding linked lists with a short description after each link. That is the entire format, and it is deliberately simple so that parsers and regular expressions can read it reliably. Who created llms.txt and why? Jeremy Howard of Answer.AI put the proposal forward in September 2024. He is a well-known figure in machine learning, and the problem he identified is real enough. Models burn tokens untangling navigation menus, cookie banners, and ad slots before they reach anything useful. Notice the word proposal, though. llms.txt is a suggested convention rather than a ratified standard, and nobody is obliged to honour it. That single distinction explains most of what follows in this article. How is llms.txt different from robots.txt and sitemap.xml? These three files sit in the same folder and do completely different jobs. The llms.txt vs robots.txt confusion comes up constantly, so here is a plain comparison. File What it does Who it speaks to Format Is it enforced robots.txt Controls which crawlers may request which parts of Search and AI crawlers Plain text Yes, honoured by all major crawlers since your site 1994 sitemap.xml Lists every URL you want Search XML Yes, actively fetched by indexed engines Google and Bing llms.txt Suggests which pages matter most and Largelanguage Markdown No, voluntary and unconfirmed summarises them models Robots.txt is the oldest of the three, introduced in 1994 as the Robots Exclusion Protocol. Thirty years of adoption is why it still carries real weight, and it is worth remembering that llms.txt is roughly two years old by comparison. The llms.txt vs sitemap.xml distinction matters more than most articles admit. Your sitemap gets discovered automatically, because search engines look for it at a known path or find it referenced inside robots.txt, then use it to guide crawling and manage crawl budget on larger sites. Nothing pulls llms.txt in the same way. It has to be requested deliberately. All three can live in your root directory together. Adding one never replaces another. Do ChatGPT, Claude, and Gemini actually read llms.txt? No, not in any confirmed way. As of mid-2026, no major provider has formally adopted llms.txt as part of its crawler behaviour, and the server log evidence points in the same direction. This is the part worth reading twice, because it contradicts most published advice on the subject. What the adoption data shows Ahrefs ran the largest study available on this. They analysed 137,000 domains and found that 28 percent published a valid llms.txt file, which works out to 38,360 sites. More than one in four. On adoption alone, the file looks like it is winning. Then they checked whether anything read those files. Of the roughly 38,000 domains with a valid file, 97 percent received zero requests for it during May 2026. No bots, no humans, nothing at all. Semrush found the same pattern at a smaller scale. They added the file to Search Engine Land in March 2025, then reviewed the server logs. Between mid August and late October 2025, the file received zero visits from Google-Extended, GPTBot, PerplexityBot, or ClaudeBot. Googlebot and Bingbot touched it a handful of times and treated it like any other page on the site. Two adoption figures worth reconciling, since you will see both quoted. Semrush cited NerdyData showing only 951 domains with an llms.txt file as of July 2025, which sits nowhere near 28 percent. The samples are not comparable. Ahrefs measured sites already running their analytics software, meaning technically minded owners who follow AI news closely, while the NerdyData count came from a broader crawl a full year earlier. Read 28 percent as an upper bound among engaged site owners rather than a figure for the web at large. Which AI companies support llms.txt? None of them formally, though the picture is not uniform. OpenAI honours robots.txt and runs separate agents for different jobs, with OAI-SearchBot handling ChatGPT search and GPTBot handling training. Neither is documented as reading llms.txt. Anthropic publishes its own llms.txt file for its documentation, which suggests openness to the idea, but has not said ClaudeBot consumes them. Google manages AI crawl behaviour through the Google-Extended token in robots.txt, with no mention of llms.txt