Google has already retired FAQ rich results as of May 7, 2026, per its FAQ structured data documentation. An Ahrefs study of nearly 1,900 pages indicated there was little to no real-world effect on citations via AI platforms using JSON-LD structured data.
What does earn citations? It is clearly stated in many sources: well-written, structurally correct, statistic-supported answers, regardless of whether they include any type of structured data or schema markup.
Why does FAQ structured data no longer guarantee AI citations? The last couple of years have been filled with my telling small business clients to place FAQ structured data onto their website. So have most agencies.
The simple reason to do so was that adding FAQ structured data takes only a couple minutes, provides you a chance at a richer search engine result, and a narrative around “being ready” for AI that makes almost everyone involved happy during the meeting with the client. However, the narrative has fallen apart and I would much rather explain what happened to you right now than wait until you spend a week preparing each of your Q/A pair formats for nothing.
As of May 7, 2026, Google quietly placed a deprecation notice into its FAQ structured data documentation. The deprecation notice states that the FAQ rich result feature will not display within Google Search beginning on that date. FAQ reporting through Search Console ends in June and API support ends in August. Only four days prior to these scheduled events, Ahrefs published information more relevant to those who still believed that FAQ structured data had relevance to the ability of being visible within AI.
Ahrefs tracked over 1,800 webpages that included JSON-LD structured data from August 2025 through March 2026, paired them against a control group of approximately 4,000 pages, and compared citation levels across Google AI Overview, AI Mode, and ChatGPT. In conclusion, there were minimal differences in citations per page among all tested platforms due to the inclusion of structured data.
In fact, per Ahrefs, Google AI Overview actually demonstrated a slight, though statistically significant, decrease of approximately 4.6% relative to the control pages. Results for both AI Mode and ChatGPT were not large enough to determine as anything other than random fluctuations. Adding structured data did not negatively affect you. It simply did not positively affect you either.
This is where the way you view the impact of this on your company needs to shift: the Ahrefs study also found that webpages which were previously cited by AI were approximately three times more likely to contain JSON-LD than webpages that were never cited by AI. Therefore, structured data correlates with citation, it does not cause it.
Businesses utilizing structured data generally have higher-quality websites, better-maintained sites, and are supported by superior content. Therefore it is those factors that contribute to citation increases, not the actual structured data markup itself.
What does get a business listed on ChatGPT & AI Overview listings?
If schema is not the method that drives the listing, then what is? The best evidence-based answer is found within academic studies vs. folklore on SEO blogs. In June 2024, researchers from Princeton University, Georgia Tech, Indian Institute of Technology (IIT) Delhi, and the Allen Institute for AI presented their first large-scale empirical study on this very same issue at the ACM SIGKDD Conference. This work introduced the concept of Generative Engine Optimization, which defines a black box methodology for enhancing content exposure in generative engine response results. Using GEO-Bench, a benchmark consisting of a diverse set of queries and actual web sources, the authors assessed their approach. One of the most significant findings from the study was that GEO methodologies can enhance exposure in generative engine responses up to 40%.
The methods that had the greatest impact were not related to technical markup. Including statistical information increased exposure by approximately 41%; referencing legitimate sources added around 28% increase; referencing outside resources added as much as 115% to exposure rates for lower ranked content. Furthermore, the researchers discovered that there was virtually no incremental gain from adding these features to content that was already ranked at #1. However, there was a substantial increase for content that was currently ranked at position #5, a respectable ranking, but not dominating.
This is certainly welcome news for Small-Medium Enterprises (SME). As mentioned earlier, you do not have to be the largest player in your market segment. You just need to have content that provides an AI system with something tangible to latch onto.
Also, quite literally, that’s where the punchline lies: SEO rewarded you for inserting keyword-stuffed phrases into invisible areas of a page. GEO rewards you for expressing a truthful and factual idea. For twenty years now I’ve watched the entire industry follow every acronym thrown by Google like dogs chasing sticks. Yet, after all those years I would have never guessed the winning strategy would be “express an opinion and support it with a statistic.” So, here we are.
In practical terms, this means that regardless of whether it is ChatGPT generating answers using live web retrieval, Perplexity generating footnoted answers, or AI Overviews providing synthesized answers, AI systems are searching for sections they can lift without modification and credibly cite. A separate study examining retrieval behavior has shown that when making citations, AI engines render and extract visible, readable HTML content at the time of citation selection. Therefore, when making citation decisions, the rendering of hidden structural data is not extracted. The content being referenced is what is actually being referenced.
Therefore, this shifts the brief for a small-medium enterprise in Japan. Regardless if your company operates as a ryokan in Kyoto, manufacturing component provider in Osaka, or professional services firm in Tokyo, instead of asking “did we add schema to this page,” the correct question should be “is this page containing a specific, self-contained statement that an AI engine could copy verbatim and still correctly reference?”
How should SMEs structure FAQ content for AI search?
The first step for SMEs to use the right format for their FAQ data in order to allow AI search to find the correct data. This is because although the structure of a question and answer (Q/A) will always have importance in terms of how a system finds relevant information based on the format of the query, the importance is now diminished due to the fact that AI can find clean Q/A content regardless if it was formatted using FAQPage schema or not.
As stated by one researcher looking into AI citation patterns, “AI systems do not favor FAQPage schema when deciding which citations to include, instead they look for clean Q/A content, whether the content has been marked up with schema or not.” This is important for understanding exactly how to improve a website’s chances for being included in an AI generated response.
One reason why clean Q/A content provides a good opportunity for inclusion in AI responses is simply because its format matches the type of query an AI is attempting to respond to. Therefore, when developing a site it is beneficial to write clear, concise questions alongside equally clear and concise answers. When done so in plain view, this allows the AI systems to easily identify and extract all of the necessary components.
A practical example of creating answers as if explaining something to someone who physically stood in front of you involves answering like an employee would explain something verbally. An example of this would be rather than stating “we provide competitive pricing,” state “the cost for our standard wedding photography package in Kyoto is $X and includes X amount of edited photographs, and delivery typically occurs within X amount of weeks.”
Specificity is crucial when developing FAQ content for inclusion in an AI generated response. This aligns with research related to citation trends and patterns associated with ChatGPT and Perplexity by a third party monitoring tool. Research indicates that websites with well developed, organized FAQ content and answer lengths ranging approximately 80 to 150 words seem to be cited more frequently than those that only provide brief, 30 word answers that previously were created primarily for Google’s old accordion style drop down menu.
What errors do you need to pay attention to while creating FAQ content for AI search?
The largest error is to treat FAQ content like an SEO check box rather than providing truly helpful services for customers. Google has consistently advised restricting the number of FAQs to no more than three to eight, and warned that having too many will dilute the relevancy of the information contained within and lead to looking spammy. If you create hypothetical questions to include on your page simply to include a keyword, you’re making the exact type of thin content that Google wants to prevent through their restrictions introduced in both 2023 and 2026.
Another mistake is focusing on the schema rather than the content. According to Ahrefs’ reporting, schema is used much more frequently on pages that have been mentioned by AI, which has caused some marketers to believe that using schema increases visibility, when there is a much better chance that companies who invest time into developing strong content are also going to be investing in establishing more authority in multiple other ways. Do not confuse correlation with causation. Spend the hour you were going to waste trying to debug JSON-LD syntax instead writing another clear and well-researched answer to add to your existing page, a conclusion Search Engine Journal reached in its own coverage of the same study.
A third mistake, and one that is specifically related to longer pieces of content that are either multi-page or based on templates, is duplicating the same response with slightly different wording throughout five different pages. Comprehensive and non-redundant coverage of a subject matter is rewarded by AI systems. Five thin, nearly identical FAQ pages do nothing for you, whereas one complete piece of content that provides a clear and direct answer to the question will provide value to a customer.
What should you accomplish this week?
Review your five least successful product/service pages and evaluate each according to this criteria: if an AI system directly referenced this page, would the reference be something useful to the customer, or simply generic filler that could be applied equally well by any competitor? Write new versions of those pages with a unique number, name of a process, or actual date associated with the information. You don’t need to remove any current FAQPage schema that you have implemented, but keep in mind that implementing new schema isn’t going to increase your ability to grow, and that energy is better focused on improving the quality of your answers, since that is where the real research points.
Frequently Asked Questions
Question: Does FAQ schema still help with SEO in 2026?
Answer: Not as a ranking factor, and not for rich results on most sites. FAQ rich results have been unavailable to the vast majority of websites since Google restricted them to government and health sites in 2023, and Google removed the feature entirely for everyone on May 7, 2026. Keeping the markup won’t hurt you, but it no longer produces any visible search benefit.
Question: Should I remove FAQ schema from my website?
Answer: No, not unless it’s inaccurate or references content that no longer exists on the page. Google has confirmed the markup can remain in place without causing problems, and other crawlers may still use it to understand your content, even though it no longer drives a Google rich result or, per Ahrefs’ 2026 study, a measurable AI citation increase on its own.
Question: What actually gets a small business cited by ChatGPT?
Answer: Clear, specific, statistics-backed answers written in plain visible text, not hidden markup. Academic research from Princeton and Georgia Tech found that adding named statistics and citing credible sources can lift AI visibility by 40% or more, while a 2026 Ahrefs study found schema markup alone produced no such lift.
Question: How many FAQ questions should an SME include on a service page?
Answer: Between three and eight, and only questions your actual customers ask. Google explicitly warns that padding a page with unnecessary questions to insert keywords looks spammy to both search engines and AI systems, and it dilutes the pages that are genuinely useful.
Q: Does this apply to ChatGPT the same way it applies to Google AI Overviews?
A: Mostly yes, with one distinction. Schema has limited effect on ChatGPT’s core training data, but it can factor into ChatGPT’s live web retrieval mode alongside dozens of other signals. Across every platform tested by Ahrefs, including ChatGPT, adding schema alone did not produce a meaningful citation increase, which is why content quality remains the more reliable lever regardless of which AI system a customer is using.
Conclusion
FAQ schema for SMEs was never a bad idea. It just stopped being the thing that matters most, and the businesses still treating it as their whole AI visibility strategy in mid-2026 are optimizing for a rich result that Google has already turned off. The businesses actually getting quoted in ChatGPT and Google AI Overviews right now are the ones answering real customer questions with real numbers, in plain readable text, whether or not a single line of JSON-LD sits underneath it. Fix the answer first. The markup was always secondary.
If you’d like help auditing your service pages against what’s actually driving AI citations in 2026, Ongito can walk through it with you.