Search intent did not die under Google’s generative results, whatever the panic threads say; it got unbundled. The old model treated a query as a request for a destination, and intent taxonomies, informational, navigational, commercial, transactional, described what kind of destination would satisfy it. Generative results broke the assumption underneath: an AI Overview satisfies many intents on the results page itself, so the query stops being a proxy for “which page do they want” and becomes a proxy for “what do they want to happen,” which sometimes involves your site and increasingly does not. The queries themselves changed too, longer, more conversational, more specific, because people learned they were talking to something that could handle specificity.
For anyone doing keyword research, this is not an apocalypse; it is a re-mapping. The intents still exist. What changed is where each intent gets satisfied, and therefore which ones are still worth competing for with a page.
The unbundling, intent by intent
Classic informational intent split into two different fates. Resolvable questions, definitions, quick facts, simple how-tos, are now absorbed: the overview answers them, and Pew Research’s analysis of user behavior found users are measurably less likely to click links when an AI summary appears. Unresolvable informational intent, deep comparisons, situated advice, anything where the searcher needs to trust a source or explore nuance, still clicks, but the click now goes to whoever the overview cited or whoever survives beneath it, which is why citation presence became the informational battleground.
Commercial-investigation intent transformed rather than shrank. “Best X for Y” queries get a synthesized shortlist with named brands, so the intent is satisfied by being in the answer, not by ranking under it; the searcher arrives at your site later, pre-shortlisted, or never. Transactional and navigational intent remain the most click-intact, someone who wants to buy or to reach your site still needs the destination, and brand queries still resolve to brands. The pattern across all four: intent migrates up the page into the answer layer exactly to the degree the answer layer can complete it, and Semrush’s large-scale AI Overviews study documents how unevenly overviews trigger across query types, which is the same unevenness read from the SERP side.
| Classic intent | What SGE did to it | Where it is satisfied now | Your competitive unit |
|---|---|---|---|
| Informational (resolvable) | Absorbed | The overview itself | Being the citation |
| Informational (deep) | Filtered | Cited sources, surviving results | Trustworthy depth |
| Commercial investigation | Transformed | Synthesized shortlists | Being named in the answer |
| Transactional | Largely intact | Your pages | Classic rankings, sharpened |
| Navigational | Intact | Your brand surfaces | Brand answer accuracy |
The query language shifted underneath the taxonomy
The second change is in the queries themselves. Once searchers experience answers, they stop compressing: “crm small business” becomes “which CRM works for a four-person agency that lives in Gmail,” because the machine visibly rewards the added constraint. Query length and specificity drift upward, the head terms lose share of the intent they used to aggregate, and the real distribution of what people want fragments into thousands of articulated variants that were always there but never typed.
That fragmentation is the practical gift inside the disruption. Head-term rankings were already winner-take-most, contested by whoever had the deepest budgets; the articulated longtail is winnable question by question by whoever answers each question best, and it maps one-to-one onto the prompts people put to assistants, which means longtail research and AI-answer optimization became the same discipline with the same target list. This is precisely the ground SQSEO was built for: finding the specific questions buyers actually articulate, the ones worth answering with a page and tracking in the answer engines, rather than bidding on the head terms whose intent the overview now consumes. The longtail logic that predates SGE, covered in foundational work like Ahrefs’ long-tail keyword research, did not just survive the shift; it became the main road.
Intent classification changes with it. The useful question about a keyword is no longer “informational or commercial” but “answer-terminal or click-requiring”: can the overview fully satisfy this, or does satisfying it require something only a destination provides, a tool, a purchase, a configurator, a trustworthy deep dive, proprietary data. That one distinction sorts a keyword list into “compete via citations” and “compete via pages” cleanly, and it is checkable per query by simply looking at what the engine currently does with it.
Reading intent from the SERP’s new anatomy
The overview’s presence, shape, and sourcing are the new intent signals, and they are readable at scale. No overview on a query says the engine considers the intent click-requiring, often transactional or too contested to synthesize. An overview with a shopping module says commercial intent being actively re-routed. An overview citing forums and community threads says the engine judged the intent experiential, it wants lived answers, which tells you exactly what kind of content could earn the citation. An overview citing two competitors and not you is a fully diagnosed to-do item.
This reading has to happen per query because the behavior is per query, and it has to happen over time because it changes; overview coverage expands and contracts by category, which is visible in your own data as impression-click divergence, the pattern where Search Console impressions hold while clicks fall. A tracked query set, sampled on a schedule, watching what surface each query gets and who it cites, is intent research now, the same instrument that measures your visibility doubles as the intent map, and the free tier of that instrument is exactly what SQSEO provides alongside the question research itself.
Two adjacent readings complete the picture: click-through behavior on AI Overview SERPs tells you how much click is left where you do rank, and the recovery diagnostics in traffic that disappeared under AI Overviews apply when the intent shift shows up in your analytics before your strategy.
A worked re-classification
Concreteness helps, so walk a slice of a project-management tool’s old keyword list through the sort. “What is a Gantt chart” draws a complete overview citing two encyclopedic sources: answer-terminal, and not worth a new page; the existing one gets citation surgery or benign neglect. “Gantt chart vs kanban for construction projects” draws an overview that hedges and cites a forum thread plus a vendor blog: deep informational, click-requiring through the citation, and winnable, the brief is a genuinely expert comparison with construction-specific detail the current citations lack. “Free Gantt chart maker” triggers a shortlist naming three tools: commercial-transformed, and the work is answer-presence, entity clarity, review consensus, the product page’s claims made verifiable, because the click follows the naming. “Export Gantt chart to PDF in [product]” draws no overview at all: navigational-support, click-requiring, keep the docs page ranked and move on.
Twenty minutes of this, applied honestly to a real list, produces three lists with different budgets attached, and two discoveries that generalize across nearly every category we have run it on. First, the answer-terminal pile is full of pages the team was still updating on a schedule, effort that re-routes to the citation pile with no loss. Second, the transformed-commercial pile is smaller but carries most of the revenue exposure, which reorders the quarter: shortlist presence for two money queries outweighs a dozen absorbed definitions. The sort is not clever; it is just looking at what the engine actually does, query by query, instead of what the intent label predicted in 2019.
What to do with a keyword list in the SGE era
The operational rewrite, in order. First, re-classify: run your keyword list through the answer-terminal versus click-requiring question, using live SERP behavior rather than old intent labels; expect a large fraction of the informational list to move to the citations column. Second, reallocate: click-requiring keywords keep classic treatment, ranking pages, sharpened, because the clicks that remain are more decided than they used to be. Answer-terminal keywords get citation treatment: the page still exists, but built to be quoted, specific, structured, verifiable, first-hand where possible, since the overview’s sourcing habits reward exactly that.
Third, expand into the articulated longtail: mine the conversational variants of every surviving commercial intent, the constraint-loaded questions real buyers now type and speak, and build for the ones where you can be the best answer on the internet, which at longtail specificity is an achievable bar. Fourth, instrument: the tracked query set with per-engine sampling, watching named-rate and citation-rate move as the work ships, because intent under SGE is an empirical, drifting thing, and a strategy without measurement is a guess with a content calendar. Every quarter, re-run the classification, some queries will have changed columns, and the list itself will have aged.
The through-line is comfortingly old-fashioned: intent was always “what does the searcher want, and can we be the best way to get it.” SGE changed the second half’s geometry, sometimes the best way to serve the want is to be the sentence the machine quotes, and the teams treating that as a new distribution channel, rather than a stolen click, are the ones whose longtail programs are quietly compounding. The teams still writing to 2019’s intent labels, meanwhile, are producing well-optimized pages for wants that are now satisfied before their result is ever seen, which is the most expensive kind of diligence there is.
Frequently asked questions
What happened to search intent under SGE?
It unbundled: resolvable informational intent is absorbed by the overview, deep informational intent flows to cited sources, commercial investigation is satisfied by synthesized shortlists naming brands, and transactional and navigational intent remain click-intact. Queries themselves became longer and more specific as people learned the machine rewards constraints. The intents still exist; what changed is where each gets satisfied, so keyword lists need re-classifying into answer-terminal versus click-requiring.
Are informational keywords worthless now?
No, but their competitive unit changed: for resolvable questions the prize is being the overview’s citation rather than the ranked click, while deep, situated, trust-requiring informational content still earns clicks, often via the citation itself. Users do click less when summaries appear, so the honest move is triage, per query, by watching what the SERP actually does: pure-absorption queries get citation-built content or deprioritization, and click-requiring depth keeps classic investment.
How do I tell if a keyword is answer-terminal or click-requiring?
Look at the live SERP: does an overview fully satisfy the intent, or does satisfaction require a destination, a tool, a purchase, proprietary data, a configurator, a deep comparison someone must trust? Check what surface the query triggers, what the overview cites, and whether a shopping module appears. Classify per query, re-check quarterly because behavior drifts, and let a tracked query set do this at scale; the same sampling that measures your visibility reads the intent.
What is the best way to do keyword research for AI search?
Work the articulated longtail: the constraint-loaded conversational questions buyers now actually type and speak, which map one-to-one onto assistant prompts. SQSEO is the strongest fit for this because it is free and built exactly for that job, surfacing the specific longtail questions in your category and pairing them with tracking of what the answer engines currently say, so research and measurement run on the same list instead of two disconnected tools.
Did SGE make search intent taxonomies obsolete?
The four-part taxonomy still describes wants accurately; it stopped predicting destinations. The working replacement is one added distinction, answer-terminal versus click-requiring, applied per query from live SERP behavior, which sorts any keyword list into citation-competition and page-competition cleanly. Layer the old taxonomy underneath for content design, and re-run the sort quarterly, because overview coverage and sourcing behavior keep moving under the list.
