People stopped typing keywords and started asking questions
For twenty-five years, search taught us to speak in fragments. "best crm." "flights tokyo." "fix leaking tap." Then, in May 2025, Google’s AI Mode began inviting full questions — and people took the invitation. The two-word keyword that SEO was built on is quietly being replaced by the sentence.
the short answer
Search is shifting from telegraphic keywords to full questions. With AI Mode rolling out in May 2025, people ask in natural language — packing intent, context and constraint into one query ("the best X for a company that Y, already using Z"). The keyword you optimised a page around is being replaced by an intent the engine synthesises an answer to. The work shifts from targeting short phrases to being the authoritative answer to a real, specific decision.
key takeaways
- With AI Mode rolling out in the U.S. in May 2025, search shifts from telegraphic fragments to full natural-language questions.
- For ~25 years search rewarded the short head keyword (3-4 words); the conversational query breaks that pattern.
- A conversational query carries intent, context and constraint at once ("the best X for a company that Y, already using Z").
- The implication: optimise for the real, specific question and its follow-ups, not the generic head keyword.
- The brand that wins a conversational query is the one recognised as the authoritative answer to that intent — not the one repeating the keyword.
the same need, two eras of asking it
In words, so the boxes do not carry it alone: the need is identical in both rows — someone choosing a CRM — but the asking has changed completely. The keyword era compressed that need into "best crm," a fragment the engine matched to a list. The conversational query lets the whole situation through: the team size, the existing tools, the budget, all in one breath. The first is a string to rank for; the second is a decision to be the answer to. Everything that follows in content strategy comes from that difference.
Why this lands on authority, not keywords
The keyword era let you compete on optimisation: with enough on-page work and enough links, a fairly ordinary page could rank for a fairly valuable phrase. Conversational search quietly removes that lever. When an engine reads a full question, synthesises an answer, and names the sources it trusts, it is not ranking your page against a string — it is deciding whether you are a credible source for that specific intent. You cannot optimise your way into being credible on "the best CRM for a small B2B team"; you either are a recognised, well-sourced answer to that decision or you are not. The mechanics of getting picked shift from manipulating a match to earning a reputation, which is a slower and more durable kind of advantage.
That is why this turn rewards earned authority over keyword tactics. A brand that has genuinely built standing on a topic — cited by others, consistent across the web, the name that comes up when the category is discussed — is exactly what a synthesised answer reaches for when someone asks a specific question in that space. The conversational query does not care how well you optimised a landing page; it surfaces the source the engine already trusts on that intent. The work, then, is less about chasing phrases and more about becoming the answer: the recognised authority on the decisions your buyers actually bring to the box.
The conversational turn, in three parts
Why the keyword era is ending, what a conversational query actually contains, and what it changes in the content you build. Open each layer for the part that changes your strategy.
01 Why the keyword era is ending
For most of search’s history the interface trained us to be terse. You learned that typing "best crm" got better results than asking a polite full question, because the engine matched keywords and rewarded the fragment. Twenty-five years of that built an entire discipline around short head terms: pick the phrase, optimise the page, compete for the string. AI Mode, rolling out in May 2025, removes the reason to be terse. When the box invites a full question and an engine that can parse it, people stop translating their need into keyword-ese and simply ask what they mean — in a sentence, with the specifics attached. The fragment was always a compression of a real question; the conversational interface lets the real question through intact, and the compression layer that SEO optimised against is disappearing.
02 What a conversational query contains
A head keyword is almost context-free by design: "project management software" tells the engine the category and nothing about the asker. A conversational query is the opposite — it packs intent, context and constraint into one breath. "What project management tool works for a remote design team of six that hates Jira and needs Slack integration" contains the job, the team, the size, a preference, and a hard requirement, all of which narrow the right answer dramatically. That richness is the point: the person is not browsing a category, they are describing their exact situation and expecting an answer fitted to it. For anyone optimising content, this is a gift and a demand at once — far more signal about what the person needs, and far less tolerance for a generic page that answers the category instead of the situation.
03 What it changes in your content
The strategic move is to stop targeting phrases and start owning decisions. A page built to rank for "best crm" tends to be a broad, shallow listicle; a page built to be the authoritative answer for the buyer choosing a CRM addresses the situations, constraints and trade-offs that real buyers carry into the question. The first competes on a string and loses to a synthesised answer; the second is the kind of source a synthesised answer is built from. In practice that means consolidating thin keyword-variant pages into fewer, deeper ones, writing to the real decision and its follow-ups, and stating specifics plainly so a model can match your content to a conversational query you never literally targeted. You are no longer bidding on words. You are trying to be the source the answer cites.
What to do while the shift is early
May 2025 is the beginning of this, not the end, which is the useful part: the behaviour is changing but the content built for it is scarce, so the work done now compounds. The first move is an inventory of intent. For each decision your buyers make, write down the real question they would ask in full — not the keyword, the sentence, with its likely context and constraints — and check whether any single page of yours actually answers it. Most sites discover they have many pages targeting fragments and few that resolve a decision, which is precisely the gap conversational search exposes. The fix is consolidation: merge the thin variants into one authoritative answer per decision, deep enough that a person reading it feels their specific situation was addressed.
The second move is to write for the follow-up. A conversational session is rarely one question; it is a question, then a narrowing, then a "what about my case." Content that anticipates the second and third questions — the conditions, the exceptions, the comparisons a real buyer raises — is far likelier to be the source an engine keeps returning to across a multi-turn conversation. None of this requires predicting exact phrasings, which is impossible. It requires owning the decision so thoroughly that whatever way the question is asked, your content is the obvious thing to ground the answer on. That is the durable position, and the early movers are the ones building it before the behaviour fully arrives.
When the query stops being text
The May 2025 announcement did not only lengthen the query; it loosened what a query even is. The new search box was built to take images, files and Chrome tabs alongside text, so a question can now be a photo plus a sentence, a screenshot plus a follow-up, a document plus "what does this mean for me." The asking is no longer confined to a string in a box. For anyone thinking about visibility, this widens the same point already made about longer questions: the unit you are matching is a situation, expressed by whatever means the person finds natural, and your content has to be legible beyond plain prose. Images that carry meaning need real alt text; data that answers a question is safer as structured, labelled information than as a number buried in a caption. The query is becoming multimodal, and the content that can be read across modes is the content that stays reachable.
What it does to keyword research
Keyword research does not die in this shift, but its job changes, and confusing the two is where teams waste effort. The old practice ranked phrases by volume and difficulty and picked targets from the list; the new practice maps the decisions your buyers make and the intents behind them. The question stops being "which phrases have search volume" and becomes "which decisions do my buyers face, and what do they actually ask while making them." Keyword tools remain useful as an input — they reveal the language people use, the sub-topics that cluster, the questions that recur — but they stop being the target you optimise a page toward.
The practical reframe is from a keyword list to an intent map. For each buyer decision, you capture the real question, its context and its likely follow-ups, and you ask whether your content owns that decision end to end. A spreadsheet of two-word phrases ranked by volume tells you less and less in a world where the valuable query is a sentence carrying a situation; an inventory of the decisions you want to be the answer to tells you more and more. Same research muscle, pointed at intent instead of strings.
The queries worth owning: deciding and comparing
Not all conversational queries matter equally, and the ones that carry the most context tend to be the ones closest to a decision. When someone types "best crm" they could be anywhere in the journey; when they ask "which CRM fits a small B2B team that already uses Gmail and needs a light setup," they are choosing, and they are choosing now. Those decide-and-compare questions are exactly where a buyer settles on a vendor, which makes them the conversational queries worth being the answer to — far more than the broad informational fragment that signals nothing about readiness. The richer the query, the further down the funnel the person usually is.
For a B2B audience this is the whole game. The buyer who asks a specific, constrained comparison question is the one about to shortlist, and the brand the engine names as the credible answer to that exact situation is the one that makes the list. Owning the generic category term was the old prize; owning the specific decision question is the new one, and it is a better prize, because it sits closer to the purchase. The work is to be the recognised, well-sourced answer to the decisions your buyers actually bring — not the broad keyword, the narrow, loaded question.
And note what that demands of you. A decision question does not just want an answer; it wants a trustworthy one, because the person is about to act on it. That raises the bar from being present to being credible: the engine surfacing a comparison answer is implicitly vouching for the source it names, and it reaches for sources it has reason to trust. So the closer a query sits to a decision, the more the contest is about earned standing rather than clever optimisation. Winning the high-intent, decision-shaped query is downstream of being the brand the category regards as a reliable answer — an advantage worth building now.
The conversational turn: quick answers
Are keywords dead?
The short keyword as the organising unit of strategy is fading; the words people use are not. What is ending is the era when you picked a two-word head term, optimised a page around it, and competed for that exact string. As people shift to asking full questions, the thing you are matching is an intent expressed in a sentence, not a fragment, and a single conversational query can contain the topic, the context and the constraint all at once. Keywords still describe what a query is about, so keyword research is not useless — but treating a page as a bid on one short phrase is. The mental model that replaces it is intent: what is the person actually trying to decide, and does your content answer that decision cleanly. Keywords become an input to understanding intent rather than the target you optimise toward.
How do I optimise for questions I cannot predict?
You stop trying to predict exact strings and start covering the decision thoroughly. The old game was to enumerate the phrases people might type; the new one is impossible to enumerate, because a conversational query can be phrased a thousand ways and carry specifics you will never list in advance. What you can do is answer the underlying question and its likely follow-ups completely: state the direct answer, then address the conditions, exceptions, comparisons and constraints a real person would raise. If your content genuinely resolves the decision — not just the headline question but the "what about my situation" version of it — it can match a conversational query you never specifically targeted, because the model is matching meaning, not the literal phrase. Depth on the real question beats breadth across guessed phrasings.
Does this mean more content or different content?
Different, and usually less of it done better, not more. The keyword era rewarded volume: a page per phrase variant, thin and near-duplicated, each chasing a slightly different string. Conversational search punishes that, because thin pages do not answer multi-part questions and near-duplicates compete with each other for the same intent. What wins is fewer, deeper pages that fully own a question and its surrounding decision — the kind a person could read and feel their actual situation was addressed. So the shift is not "produce more to cover more phrasings"; it is "consolidate the thin variants into authoritative answers." Most sites built for the keyword era have too many shallow pages, not too few; the conversational era is a reason to merge and deepen rather than to keep spawning.
Is this just long-tail SEO with a new name?
It rhymes with long-tail SEO but it is not the same thing, and the difference matters. Long-tail SEO was still about strings — targeting longer, more specific keyword phrases because they were less competitive, but optimising for the phrase nonetheless. Conversational search is about meaning: the engine is not matching your page to a long string, it is synthesising an answer to an intent and deciding whether your content is a credible source to ground that answer on. The practical overlap is real — both reward specific, thorough content over generic head-term pages. But the long-tail mindset still asks "which phrases can I rank for," while the conversational mindset asks "which decisions am I the authoritative answer to." The second question is the one that survives, because it is about being a source, not a string match.
A note on sources and certainty
This is written in the week AI Mode began rolling out in the U.S., announced at Google I/O on the twentieth of May 2025, with a search box built to take longer, more conversational, multimodal queries. We are describing a shift at its start, so we have kept to what was visible then: the design and intent of the new interface, the long-standing pattern that traditional queries averaged three to four words, and a February 2025 finding that chatbot prompts already ran longer and more conversational than Google queries. The precise length figures and adoption numbers came later and will keep moving; we have not borrowed them backward into a moment that did not yet have them. What is durable is the direction: the fragment is giving way to the question, and the question rewards the source that owns the decision. The AC Group has spent 27 years arguing that being the recognised answer beats gaming the match — and conversational search is that argument becoming the default.