Google AI Overviews and AI Mode: how to get cited

To show up in Google AI Overviews, be the clearest, most credible source Google can summarize for a specific question: answer it directly near the top, structure the page so the answer is easy to lift, and earn enough authority that Google trusts you. AI Overviews are not a separate game from SEO, they are SEO with a higher bar for clarity and credibility.

AI Overviews now sit above the classic results for a large share of informational searches. Google reads a handful of sources, writes a summary, and links to the ones it used. If you are one of those sources, you stay visible at the very top of the page. If you are not, the Overview can answer the user before they ever scroll to your link.

What an AI Overview actually is

An AI Overview is a generated summary Google places at the top of the results for some queries, built from multiple sources and linked back to them. Instead of only ranking pages, Google retrieves several, synthesizes an answer, and cites the ones it leaned on. It is the same retrieve-read-synthesize-cite pattern behind answer engine optimization generally, applied inside Google's own results page. All of this assumes Googlebot can actually reach the page: if you run Cloudflare and have blocked AI crawlers at the edge, check that the rule didn't catch Googlebot too, since Cloudflare's content signals policy can quietly do exactly that. Google also ships a Search Console setting for this, an AI content opt-out toggle that excludes a property from AI Overviews and AI Mode without touching rankings.

That changes the goal. You are no longer only competing for a blue-link position a human scans. You are competing to be a passage Google trusts enough to summarize and attribute.

The scale of the shift is the part worth internalising. SparkToro, using Similarweb panel data, found 68.01% of Google searches ended without a click between January and April 2026, up from 60.45% in 2024 and 49% in 2019. Rand Fishkin's conclusion is bleak and probably right: "there's not much point (nor any hope) of fighting back by simply getting better at SEO."

And the link between ranking and citation has genuinely come apart. Ahrefs, across 4 million AI Overview URLs from 863,000 keyword SERPs, found only 38% of AI Overview citations come from pages ranking in the top 10, down from about 76% in its July 2025 study, with 31.2% coming from positions 11 to 100 and 31.0% from beyond the top 100. Louise Linehan of Ahrefs put it directly: "AI Overviews have shifted, and ranking for the user's exact query is no longer a guarantee of visibility." BrightEdge's own panel puts the top-10 slice at 16.7%, while measuring overall overlap between AI Overview citations and organic rankings rising from 32.3% to 54.5% over the 16 months to September 2025. The two sources disagree on the headline; treat the exact figure as contested and the direction as settled.

AI Overviews and AI Mode are different surfaces

Both are AI answers from Google, they cite mostly different pages, and optimising for one does not reliably win the other.

Ahrefs measured 13.7% URL overlap between the two across 540,000 query pairs, rising to only 16.3% among top-three citations, while the answers themselves were 86% semantically similar. SE Ranking found 10.7% URL overlap and 16% domain overlap across 4,281 queries. Victorious, across 1,540 queries, found 23% of domains appeared in both and 77% appeared in only one, with AI Mode citing about 9 domains per query against 7.7 for AI Overviews.

Same answer, different sources. That is unusually good news: the citation space is genuinely open, because being the best-ranked page is not what is deciding it.

One date matters. At Google I/O on 19 May 2026 Google merged the two experiences more tightly and made Gemini 3.5 Flash the default model in AI Mode globally, reporting that AI Mode had passed a billion monthly users. Gemini 3.6 Flash superseded it in July 2026, so treat any named default model as a snapshot.

Be sceptical of the widely-repeated claim that AI answers became Google's default view on a specific date in July 2026. We could not find a Google announcement supporting it, and Google's own I/O framing was: "You'll continue to get a range of results from Search, just like you do today." For a measured penetration figure rather than a claimed switch, Similarweb data reported by TechCrunch in July 2026 put AI Overviews on about 43% of searches.

How query fan-out changes what you write

Google's documentation is explicit that both surfaces use it: AI Overviews and AI Mode may each run a query fan-out technique, issuing multiple related searches across subtopics and data sources to develop a response. Google defines it as "a set of concurrent, related queries generated by the model to request more information and fetch additional relevant search results to address the user's query."

It is decades-old query expansion run through a generative model, and the patent trail is public: US20240289407A1 (Search with Stateful Chat), US12158907B1 (Thematic Search, which retrieves first and clusters into themes afterwards) and US11663201B2 (Generating Query Variants Using a Trained Generative Model).

The volume is the surprise. Seer Interactive ran 501 prompts through the Gemini 3 API with forced grounding and measured an average of 10.7 sub-queries per prompt, ranging from 3 to 28. That is a 78% jump on Gemini 2.5's 6.01. The average sub-query was 6.7 words, 95% had zero measurable search volume, and 26.4% included a brand name.

Sub-queries fall into eight types, drawn from the patents: equivalent, follow-up, generalization, specification, canonicalization, translation, entailment and clarification. Writers routinely cover the first two and skip follow-up, entailment and clarification, which is where the easy wins are.

What to do with it:

• Write each H2 and H3 as the literal sub-question, answered in the first two sentences.
• Make passages self-contained. Retrieval works at the passage level, roughly 100 to 300 words per Lumar. Repeat entity names instead of pronouns, because a chunk gets read without the paragraph above it.
• Do not build a page per sub-query. Google is explicit: "There's no requirement to break your content into tiny pieces for AI to better understand it," and creating separate content for every anticipated variation "violates Google's scaled content abuse spam policy." Consolidate near-duplicates. The test is whether a reader who skips a section misses information or just a rephrasing.
• Measure over weeks, not appearances. The signal that this is working is a widening set of distinct sub-queries citing one URL, not a single sighting.

Why Reddit keeps winning the citation

If you have watched Reddit take citations you expected, this is why, and some of it you cannot beat.

Google signed a content licensing deal with Reddit in February 2024, one day before Reddit's IPO filing, reported by Reuters at around $60 million a year. That is a structured data pipeline, not scraping. On 6 May 2026 Google announced AI Overviews and AI Mode would directly quote Reddit, social platforms and public forums with creator attribution: "For many searches, people are increasingly seeking out advice from others. To help you find the most helpful insights to explore further, AI responses will now include a preview of perspectives from public online discussions, social media, and other firsthand sources."

The numbers are lopsided. Tinuiti's Q1 2026 report puts Reddit at 44% of Google AI Overviews' social citations against 5% of Gemini's, and at 0.1% of all Gemini citations versus more than 5% of all ChatGPT citations. The Stacc found Reddit in about 21% of AI Overviews and present in 92.8% of potential citation opportunities across 10,000 citations. Profound, tracking August 2024 to October 2025, ranks Reddit the most-cited domain on Perplexity and second on both ChatGPT and Google AI Overviews, at 3.11% of all citations against Wikipedia's 1.35%.

What is actually being rewarded is copyable: threads are pre-formatted question-and-answer pairs, upvotes and moderation act as a continuous public quality filter, and answers carry first-hand dated specificity. Profound's read is that "Reddit is the most potent source of authentic and contextually rich consumer signals in AI search."

So show disagreement and edge cases instead of one tidy answer, trade generic advice for numbers with dates and outcomes, and make every FAQ answer name its own source rather than referring to "the same analysis" above it.

Which queries trigger Overviews

Informational and how-to queries trigger Overviews most: questions where Google can assemble a useful summary from several sources. Navigational queries (someone looking for a specific site) and clearly transactional ones trigger them less. The practical rule: if a query reads like a question someone wants explained, assume an Overview may appear, and write the page to be its source. "How much does X cost" reads exactly like that kind of question, which is why we treat pricing page SEO as its own live surface rather than a side effect of the rest of the site.

Since May 2026, some of those informational queries now surface a Reddit thread or forum comment alongside the summary, quoted directly under an "Expert Advice" label. We break down why that surface change happened and what it means for a SaaS blog in Reddit and AI Overviews.

Answer the question near the top

Lead with the answer. If the page targets "how to show up in AI Overviews," the first lines should answer it before any setup, the way this post does. Google's summary is built from passages that state something clearly and confidently. A page that opens with throat-clearing and saves the answer for paragraph six gives Google little to extract.

Keep the answer self-contained: a sentence or two that would make sense quoted on its own, because that is exactly how it may be used.

Structure so the answer is easy to lift

Write headings as the questions people ask, then answer each one directly beneath it. This is the same structure that wins featured snippets, and snippet-winning pages are strong candidates for Overview citations because they have already proven they can be summarized. Short, scannable paragraphs, clean lists where they fit, and an FAQ section that pairs real questions with direct answers all give Google clean units to pull from. Structured data sits alongside this as hygiene, not a lever; we cover which schema is actually worth shipping in schema markup for AI Overviews and AI Mode.

Back every claim with a credible, current source

Google summarizes sources it can stand behind. A specific, dated, referenced fact is safer to cite than a vague claim, because Google can attribute it with confidence. Undated stats and unsourced assertions are risky for the model to repeat, so they get passed over. Keeping facts verifiable and current is not optional here, which is why we treat fact-checking as a hard gate. Add a clear "as of" date on anything time-sensitive.

Earn the authority to be trusted

Overviews lean toward sources Google already considers credible on the topic. That credibility comes from depth and connection: thorough coverage of a subject, pages that link to each other sensibly, and a few external references that vouch for you. Building a connected cluster rather than scattered one-off posts is the same SEO for SaaS move that compounds in classic rankings, and it is what makes Google comfortable summarizing you. Strong internal linking is the cheapest part of that, and the most neglected.

What to measure once ranking stops predicting traffic

The diagnostic that matters: rank flat but clicks falling is an AI-citation problem, not an algorithm penalty. Teams spend months chasing a demotion that never happened because they only ever look at position.

Start with one comparison. Pull 90 days of GA4 sessions against GSC impressions for your top 20 pages, joined by URL. Pages holding impressions while losing sessions are being answered above the fold. Then check the freshness date on those same 20 pages, because freshness is now a stronger predictor than position: SE Ranking's AI Mode study found pages refreshed within two months average 5.0 citations per query against 3.9 for content over two years old, and that domain traffic is the strongest citation predictor for AI Mode at a SHAP value of 0.63, with sites over 1.16M monthly visitors averaging 6.4 citations against 2.4 for sites under 2,700. For ChatGPT specifically, SE Ranking found referring domains matter more than traffic.

One reporting caveat that catches people out: the GSC Generative AI report combines AI Overviews and AI Mode impressions into a single number. Given the two surfaces share only 13.7% of their URLs, that aggregate hides exactly the shift you are trying to detect. Until Google splits it, a manual prompt log or a paid tracker is the only way to see which surface you are winning.

A weekly scorecard worth keeping to five rows: AI citation presence by page, GSC Generative AI impressions, page freshness date, organic sessions against rank, and revenue by landing page. Reported traffic losses in the affected window ran broadly in the 20 to 60% range depending on category.

Why this is hard to sustain

Google is now testing paid placements inside the AI Mode version of this box, too. Conversational Discovery ads and Highlighted Answers sit directly inside AI Mode answers, which raises the cost of missing the citation in the first place: a page that isn't the source now competes with a Sponsored pick as well as with Gemini's own synthesis.

The same single-source-answer pattern is now moving to voice. OpenAI's GPT-Live launched full-duplex voice with live web search inside ChatGPT in July 2026, and it leans on the exact precedent set here: whichever page already proved it could answer a question directly and get cited is the one best positioned when a spoken answer needs a source to lean on.

Every step here is straightforward once. The difficulty is doing all of it on every page, forever: answering first, structuring for extraction, verifying each fact, dating claims, and keeping the internal links and authority signals current. That discipline is exactly what slips when a small team gets busy, and an Overview will quietly drop a source whose facts have gone stale.

That repetitive, unforgiving work is what we built Lyra to carry. She writes for extraction by default, fact-checks before shipping, structures posts with question-shaped headings and FAQs, and keeps the internal links honest, on every post, then opens each as a pull request you merge. Showing up in AI Overviews is mostly a matter of doing the fundamentals without exception, which is a job worth automating carefully.

AI Overviews reward the same fundamentals as SEO, enforced on every page: direct answers, verified facts, clean structure, real authority. Lyra writes that way by default and ships each post as a PR you review.

Try Lyra → · Talk to the founder