Services
Company
Industries
Knowledge
Projects
UK
.png){kind=small}
An impression-weighted analysis of AI Mode and AI Overview queries: GEO, AEO, visibility, citations, mentions, integrations, and attribution.
Data source. Atomic AGI's Google Search Console keyword data for AI Mode and AI Overview queries, 30-day window (April 20 – May 20, 2026). 101 visible queries, 1,285 total impressions.
What gets analysed. Non-branded queries only. Queries containing the project's own brand name (here:"Atomic AGI" or "atomicagi") were filtered out before analysis. Branded queries measure brand awareness, not category demand.
Anonymisation. We never reproduce exact long-tail queries in this report. What you see in each category section are structural patterns: generalised phrasings that capture the recurring shape of the queries, with vendor names abstracted to category labels. The intent is the empirical signal; the exact phrasing of any single user query is not.
Impression weighting. Each category's share is calculated by total impressions, not by query count. Impressions are the closest proxy we have to actual demand frequency: how often the question gets typed, across users, into AI surfaces. A category with three high-impression queries can outweigh one with twenty low-impression ones.
Statistical approach. Each category-share percentage comes with a 95% Wilson score confidence interval (e.g., "28.0% [25.6%–30.5%]"). The Wilson score is the recommended interval for binomial proportions and behaves correctly at small and extreme values, which matters here because the smallest categories sit below 1%. We treat each impression as a "vote" for the category of its parent query, and compute Wilson intervals on impression-weighted shares with n = 1,285 impressions.
Methodological precedent. The statistical-rigour approach follows the framework set out by Druck & Smith (Graphite, "Demystifying Randomness in AI", 2026). Their paper makes the case for Wilson score confidence intervals on proportion-style AI-measurement data, a position this analysis adopts directly.
Limitations. Three to acknowledge. First, impressions are clustered within queries (101 queries, 1,285 impressions, average ~13 impressions per query). The Wilson intervals as reported assume each impression is an independent observation; true uncertainty is somewhat larger because queries within a category share characteristics. The effect is small for the dominant categories but it matters at the bottom of the distribution. The smallest categories should be treated as directional, not precise. Second, the dataset is point-in-time. AI Mode and AI Overviews changed substantially in early 2026 and will continue to change. We re-run this analysis quarterly to track category-share shifts. Third, a small number of queries (~5%) sit ambiguously between categories; we placed each in the closest fit.
Statistically meaningful comparisons. When two categories' confidence intervals overlap, treat the difference as not statistically significant at this sample size. Specifically: Categories 4 (Recommendations, 9.7%) and 5 (Adjacent SEO tooling, 8.7%) overlap, and Categories 9 (Technical optimisation, 0.5%) and 10 (Attribution, 0.5%) are statistically indistinguishable. Everything else in the top tier (Categories 1, 2, 3) is cleanly separated.
This chart shows what people are actually asking in AI search, based on Atomic AGI data from AI Mode and AI Overviews. We grouped the queries into 10 intent categories to understand where demand is concentrated - from AI visibility and citations to integrations, pricing, trust, and vendor comparisons.
28% of all AI search demand sits in one place: tracking brand visibility, citations, and mentions across multiple LLMs (95% CI:25.6%-30.5%). Users have stopped asking how to rank in Google and started asking how to appear inside ChatGPT, Claude, Gemini, Perplexity, Copilot, and AI Overviews, sometimes in the same query.
Almost a quarter of queries (23.8%, CI: 21.6%-26.2%) are about integrations. Users treat AI search data as a layer of marketing operations, not a standalone curiosity. They want it inside GA4, Looker, Slack, and the rest of their existing stack.
17.4% of demand is still trust and skepticism (CI: 15.5%-19.6%). Even after two years of GEO and AEO conversation, users are actively asking whether dedicated AI search tools are real, necessary, and not hype.
Attribution queries, the thing every CMO talks about, are only 0.5% of the data (CI: 0.2%-1.0%). Users have not yet started asking AI engines whether AI mentions translate into revenue. The market is one step behind where marketing leadership thinks it is.
The phrasing of AI-native queries is structurally different from Google searches: longer, more conversational, more comparative, more operational. "Does platform X integrate with GA4 and Slack across ChatGPT, Claude, and Perplexity?" replaces "best SEO tool."
The first sign that AI search behaves differently from Google search shows up not in the metrics, but in the way users talk to it.
For two decades, search queries were compressed: nouns, keywords, modifiers. "Best CRM for small business." Four words, transactional. AI search dissolved that grammar. Users now ask AI engines complete questions, often containing systems-level context: "Are there tools that track brand mentions across ChatGPT, Claude, and Gemini that also integrate with GA4 and let me set Slack alerts?"
The change is more than verbose. It signals a different relationship with search. Users are no longer asking machines to retrieve information; they are asking them to reason about it. In the GEO and AEO category, where AI search analytics tools compete, that shift produces some of the clearest examples in any dataset.
The buyer side shift is no longer marginal. G2's 2026 AI Search Insight Report, published in April, found that 51% of B2B software buyers now start their research with an AI chatbot rather than Google, up from 29% a year earlier. 69% reported choosing a different vendor than initially planned based on AI guidance. The behaviour change has already happened at scale. The open question is what those buyers are actually typing.
This report analyses 101 impression-weighted questions captured inside Google AI Mode and AI Overviews over a 30-day window, drawn from Atomic AGI's own Google Search Console data. The dataset is small but dense, with 1,285 total impressions concentrated in a few high-signal categories, and the patterns are stable enough to produce category-level findings about how AI-native search has reshaped buyer language.
The recurring question shape:
"Are there tools that track [brand visibility / mentions / citations] across [ChatGPT, Claude, Gemini, Copilot, AI Overviews]?"
This is the largest category by impression weight and the most consequential. Users no longer talk about rankings, keywords, or backlinks. They talk about mentions, citations, answer presence, visibility, share of voice, and AI responses. The vocabulary of search itself has shifted, from positions on a results page to appearances inside a generated answer.
The queries in this category share three structural features. First, multi-engine awareness: users explicitly name three, four, or five LLMs in a single question, treating the AI search surface as a category rather than a single product. Second, an emphasis on appearance rather than rank: the language is about whether a brand shows up, not whether it ranks first. Third, a measurement framing: users are looking for tools that track these things, which implies they already accept the need to track them. The underlying assumption that "AI mentions matter for my brand" is no longer in question.
What this says about the market. The AI search measurement vocabulary is now stable enough to be widely searched. "AI visibility" and "citations" function as accepted category terms, not novelty phrases that need explanation. The buyer journey for AI search tooling now starts at "which tool tracks this" rather than "is this measurable."
For marketers. If your AI search content uses Google-SEO vocabulary (rankings, keywords, SERP), you're speaking a language the dominant 28% of buyer-intent queries have moved past. The terms that fit current demand are visibility, citation, mention, answer presence, share of voice.
The recurring question shape:
"Does this AI search platform integrate with [GA4 / Looker / Tableau / Slack / webhooks / dashboards]?"
This is the second-largest category and the most revealing about where buyers sit in their adoption journey. By the time a buyer is asking how an AI search tool plugs into the existing analytics and notification stack, the prior questions (is this measurable, are these tools real) have already been answered. Integration queries are the language of operational adoption, not exploration.
The queries cluster into three sub-patterns. First, analytics integration: GA4, Looker, Tableau, and the rest of the marketing-data layer. These queries assume AI search data will need to sit alongside organic search and paid-channel reporting. Second, notification integration: Slack, webhooks, alerting, custom triggers. The buyer expects to be told when something changes, not to discover it by logging in. Third, dashboard and API access: queries about embedding data into existing tools rather than learning a new interface.
What this says about the market. Almost a quarter of AI search demand assumes the tool is something to be embedded in a workflow, not a discovery toy. The buyer has moved past curiosity into operations. Tools that present as standalone dashboards face a structural disadvantage against tools that integrate.
For marketers. Integration capability belongs above the fold on category and feature pages, not buried in technical documentation or sales decks. If your AI search tool landing pages don't visibly answer "what does this plug into?", you're filtered out of the consideration set by 24% of buyer demand before evaluation begins.
The recurring question shape:
"Are GEO / AI visibility tools actually useful, or is this hype?"
Small in query count, only 7 distinct queries, but the third-largest category by impressions. A small number of skeptical questions are being typed by many users. This is the strongest signal in the dataset that the AI search category is past awareness but not past skepticism. Buyers know what GEO and AEO are; they are publicly asking whether the category itself is real.
The skepticism takes specific shapes. Some queries are direct: is this hype, is this real, do I actually need this. Others are comparative: can my existing SEO tool do this, why add another tool to the stack. Others are validation-seeking: are people actually using these tools, are there real case studies. None of them are unfamiliarity queries. These are buyers who have already absorbed the vocabulary and are now deciding whether to believe in it.
What this says about the market. A category that spends 17% of its demand on "is this real" is one where awareness has run ahead of evidence. Tool vendors have done the work of introducing the language. The work that remains is producing proof.
For marketers. The answer to "are these tools real?" cannot live in product copy. It needs to live in evidence assets: third-party case studies, independent benchmarks, named customer outcomes, transparent methodology. The category will not exit the skepticism phase through feature lists alone.
The recurring question shape:
"Do these platforms only track visibility, or do they suggest what to do next?"
Buyers want AI search tools that act, not just report. The category is shifting from analytics-software thinking (observe, then decide) to operational-AI-systems thinking, where the tool itself surfaces recommendations or takes action. The 10% share is not yet dominant, but it is a clear signal of where buyer expectation is heading.
Within the category, queries cluster along a spectrum. At the lighter end: surface recommendations and content-gap identification. In the middle: alerting on visibility changes and proactive issue detection. At the heavier end: workflow automation and direct content interventions. The common thread is movement away from passive dashboards and toward systems that close the loop between observation and action.
What this says about the market. Buyer expectations have been recalibrated by the broader AI category. A tool that costs the buyer time without saving any has a harder selling job than it would have had a year ago. Observability without intervention is no longer enough on its own; it is increasingly a disadvantage.
For marketers. Feature pages and demo content should emphasise what the tool does next, not just what it shows. Roadmap signals about automation and recommendation engines belong above the fold. Buyers reading category content compare on "what does this do for me," not "what does this report."
The recurring question shape:
"Does [established SEO platform] support [AI search capability]?"
This category is the bridge between traditional SEO tooling and AI search measurement. Rather than asking which AI-native tool to adopt, these buyers are asking whether the SEO stack they already use is adapting fast enough. The implicit question: do I need to add a new tool, or will the one I already pay for catch up?
The queries fall into two postures. The first asks established SEO platforms by name and probes whether they have shipped AI search features yet. The second asks the reverse: how AI-native specialist tools compare to the broader SEO platforms. Both postures share a common reluctance to commit to a separate tool stack. Buyers are looking for the path of least disruption.
What this says about the market. The competitive boundary between general SEO platforms and AI search specialists is being contested in real time. The category will likely settle into both modes co-existing, but for now, buyers are asking which side will win the consolidation, and reading vendor signalling carefully.
For marketers. Positioning depends on which side of the boundary you're on. AI-native specialist tools win by emphasising the depth gap, what they do that incumbents can't. Established SEO vendors win by emphasising adjacency, that they already sit in the buyer's stack and are extending into AI naturally. Mixing those messages dilutes both positions.
The recurring question shape:
"Are there free, affordable, or low-cost AI visibility tools?"
The pricing category reveals that AI search tooling has not yet settled into a clear price band. Queries cluster around "free," "affordable," and "low-cost," language that implies the buyer hasn't decided whether AI search tooling is a premium category (the way enterprise observability is) or a standard SEO-software commodity (the way most keyword research is). The market lacks a clear anchor.
The breadth of buyer type in this category is also notable. Pricing queries are not limited to enterprise teams shopping for procurement-grade software. Startups, agencies, freelancers, and smaller in-house marketing teams all show up. Demand is broad, but willingness to pay is unclear, and that ambiguity shows up in the query language itself.
What this says about the market. Categories without clear pricing standards typically sit in a first-mover-defines-it phase. Whichever vendor establishes the credible default price band, whether at the enterprise end or the freemium end, sets the anchor that other vendors get compared against later.
For marketers. Pricing transparency, freemium tiers, and clear ROI framing matter more here than they would in a mature category. Buyers in this 7.5% segment cannot calibrate value without visible pricing. Tools that hide pricing behind "contact sales" get filtered out before evaluation begins.
The recurring question shape:
"Is [tool A] better than [tool B] for AI search / GEO / visibility tracking?"
Small by impressions but large by query count: 10 distinct queries spread across many vendor pairs. The fragmentation is itself the signal. Buyers are running head-to-head evaluations, but the market hasn't consolidated yet, so the comparisons are scattered. A consideration set is forming, but it has not yet stabilised around a few dominant players.
The structure of these queries follows traditional consideration-stage comparison language: X vs Y, alternatives to Z, best AI search tool for [use case]. The patterns are familiar from any maturing software category. What's different here is the absence of any clearly dominant vendor anchoring the comparisons. The comparison surface is still wide open.
What this says about the market. The vendor set is fluid. Buyers are aware of multiple tools and are willing to compare, but no single name is yet pulling the majority of mindshare. This is the window in which positioning, comparison content, and category-defining language compound fastest. Once the consideration set narrows, the window closes.
For marketers. Comparison content (X vs Y pages, alternatives-to-X pages, head-to-head feature tables) is the dominant AI search citation surface during this window. Brands that don't appear in comparison pages don't appear in the next round of buyer consideration. Investing in comparison content is one of the most valuable activities available in a forming category.
The recurring question shape:
"Does this support [SSO / SOC-2 / onboarding / logging / permissions / historical data]?"
A small category by impressions, but its presence is the signal. Six distinct queries asking enterprise-software evaluation questions of AI search tools means a measurable portion of buyers are starting to apply the same governance lens they apply to any enterprise SaaS purchase. The category is moving upstream toward procurement-grade evaluation.
The queries are practical and familiar: security certifications, single sign-on, audit logs, user permissions, historical data retention, onboarding processes. None of them are AI-specific. They are the standard checklist of an enterprise IT evaluation, being applied to a category that, two years ago, was being adopted via individual seats and credit cards.
What this says about the market. The buyer profile is expanding. Beyond the individual marketer or growth lead piloting a tool, enterprise procurement is starting to evaluate AI search tooling as a real software purchase. The category is on a trajectory similar to any maturing SaaS category: early adopters, then teams, the enterprise.
For marketers. Compliance, security, and onboarding documentation belong on the marketing site, not buried in a sales conversation. Trust pages, security pages, and procurement-readiness content are increasingly the surface that lets a buyer pass the procurement filter. Enterprise readiness content is no longer a sales-deck-only asset.
The recurring question shape:
"What are the [technical signals / AI-native processes / structural requirements] for better AI search performance?"
A very small category: only three distinct queries and six impressions, sitting at the bottom of the distribution alongside attribution. The smallness is itself the finding. Buyers are not yet asking the technical "how does this work under the hood" questions at meaningful scale. The market is still at the tools-and-tracking layer, not the optimisation-mechanics layer.
The questions that do appear are structural rather than tactical. What signals do LLMs use to surface content, what formats get cited, what makes a page AI-discoverable. They sound like the early Google-SEO questions of two decades ago, but at a fraction of the demand share, because the buyer population that needs technical depth in AI search is still small.
What this says about the market. Technical content has less buyer pull than tool-evaluation content right now. The optimisation-mechanics conversation is happening inside a small specialist segment. As the category matures and more buyers move past the "which tool" question into the "how do I get cited more"question, this category will likely grow.
For marketers. Technical-depth content is a future bet, not a current high-value investment. Building it now positions a brand for when the category matures, but expecting it to drive immediate traffic against current demand patterns is a misread.
The recurring question shape:
"Can AI mentions / citations be tied to [conversions / revenue / pipeline]?"
The surprise finding of the dataset. Just two distinct queries and six impressions, statistically indistinguishable from technical optimisation, and the joint-smallest category at 0.5%. Yet attribution is the single topic most discussed in marketing-leadership content about AI search. The gap between what leadership debates and what buyers ask is unusually wide here.
The two queries in this category are direct, not exploratory. Can AI citations be linked to revenue. Can AI mentions be tied to conversions. These are the questions of buyers who have already accepted that AI search visibility exists and now want to know whether it produces measurable business outcomes. The smallness of the category indicates that this stage of the buyer journey has been reached by very few people so far.
What this says about the market. Buyers move through a roughly linear progression: awareness, measurability, integration, optimisation, attribution. The current centre of demand sits in the measurability-and-integration phase. The attribution phase is real, but it is still a small fraction of the active buyer population.
For marketers. Marketing leadership is debating where buyers will be in 12 months. Buyers are asking about where they are today. Investing heavily in attribution-proof content right now is a phase early. Investing in visibility, integration, and trust content matches where current demand actually sits. The attribution category will grow, and is worth tracking quarterly, but is not yet where the bulk of demand lives.
The category-share picture tells a clear story about where the GEO/AEO market is.
The four dominant categories (visibility 28%, integrations 24%, trust 17%, and recommendations 10%) together account for 79% of impressions. The first three are statistically separated from each other and from the rest of the distribution; their confidence intervals don't overlap with anything below. The market is asking: can I measure this, can I integrate it, is it real, does it act on what it sees? Those are the questions of a category exiting the awareness phase and entering operational adoption.
The three smallest categories (enterprise readiness 1.2%, technical optimisation 0.5%, and attribution 0.5%) together account for under 2.5% of impressions. Technical optimisation and attribution are statistically indistinguishable from each other; both are clearly below enterprise readiness. These are the questions of mature categories, and they aren't yet being asked of AI engines at meaningful scale. That gap is itself the finding: AI search is past category-validation but well before enterprise-procurement maturity, and the demand for business-impact proof lags marketing-leadership conversation by a phase.
Reading the data across categories, three structural shifts emerge.
The vocabulary of search has changed. Rankings, keywords, and backlinks have been replaced by visibility, citations, mentions, and answer presence. This shift is now stable; the new terms function as accepted category vocabulary, not novelty phrasings.
Search is no longer single-engine. Users routinely name three to five LLMs in a single query. The multi-engine assumption is baked into how queries get phrased, which means AI search tooling that works for only one platform looks structurally undersized.
Buyers expect AI search to be operational, not observational. Integration and recommendation queries together account for over a third of demand. Tools that only observe, without integrating into existing stacks or surfacing next-action recommendations, are increasingly out of step with how the dominant 33% of buyer-intent queries are framed.
The structural difference between traditional SEO queries and AI-native queries is sharper than the topical difference. Traditional SEO searches are compressed, noun-based, and keyword-first. AI-native searches are conversational, comparative, scenario-based, and systems-oriented.
A traditional SEO query for the same intent might read: "best AI search tool." Four words, transactional.
The AI-native equivalent, visible across multiple queries in this dataset, reads more like: "Are there tools that track brand mentions across ChatGPT, Claude, and Gemini that also integrate with GA4 and let me set Slack alerts?"
The shift is not just longer. It encodes assumptions: the tool exists, the platforms are named, integration is required, alerts are expected. Each AI-native query is partially a specification, partially a request.
Across the dataset, the strongest recurring template is:
"Does [type of platform] help with [specific AI visibility problem] across [specific AI engines] and connect to [existing workflow or tool]?"
That template appears, in varying forms, across multiple categories. It is, for now, the clearest signature of how AI search has reshaped buyer side query language.
The pattern is unlikely to revert. Pew Research's March 2026 survey found that 58% of US adults under 30 have used ChatGPT (up from 33% in 2023), and 64% of US teens aged 13-17 have used AI chatbots; 28% of them daily. The cohort whose default search behaviour was shaped by Google is being followed by one whose default is conversational AI. The query grammar documented in this study is the entry point of a more durable shift, not a transient artefact.
A useful framing first. Gartner's 2026 CMO Survey ranked brand as the #1 marketing priority for the year. AI search ranked #17. The data in this report is from a category most CMOs aren't yet prioritising, which is exactly why the buyer side signal here is worth reading early. Teams that move while leadership is still focused elsewhere set the vocabulary the rest of the market eventually adopts.
Three concrete actions for marketing teams reading this report:
This is the first of a monthly research series on AI-native search behaviour. We publish a new impression-weighted analysis on the first Monday of every month, with quarterly cross-vertical syntheses in October, January, and May.
Next month: the same impression-weighted linguistic-pattern analysis, applied to one of our platform's macro-verticals (fintech, SaaS, or web3), drawn from a much larger pool of platform projects.
Subscribe at omnius.com/research/subscribe to get each issue when it ships.
For citation: Omnius Research, "What People Ask AI Search About AI Search," June 22, 2026. Methodology derives from Druck & Smith, "Demystifying Randomness in AI" (Graphite, 2026). Dataset: Atomic AGI Google Search Console AI Mode and AI Overview queries, 30-day window (April 20 - May 20, 2026). 101 non-branded queries, 1,285 impressions, 10 impression-weighted intent categories.
Omnius is a B2B SEO & GEO agency; partnering up exclusively with SaaS, Fintech & AI companies. The result? Compounding growth made through organic positioning everywhere people search for information, including both Google & LLM search engines.
