# Intent-based interaction

> From intention to result: search based on intent, not keywords, and what it means for content producers.

Classic search required translating a need into a query, exploring results, refining terms, and comparing sources. GenAI overturns the model: the user starts from their own case ("I have this problem, in this context, with these constraints") and expects an already calibrated answer. It is search based on intent, not on keywords.

The [NN/g research on information seeking](https://www.nngroup.com/articles/ai-information-seeking-keyword-foraging/) shows that AI reduces keyword foraging (the trial-and-error work needed to find effective search terms) and handles requests with multiple simultaneous constraints — budget, timing and location — especially when traditional search requires repeated reformulation.

How little it resembles keyword search has been measured: asking 142 people to formulate a prompt for the exact same need, **the average semantic similarity between their requests was 0.081** — almost nothing. No one reduces intent to two or three keywords; everyone writes idiosyncratically, verbosely, and very specifically (SparkToro/Gumshoe, January 2026). Good news for content producers: the model still captures intent beneath the phrasing, and those 142 prompts produced answers with the same narrow set of brands. The work, however, definitively shifts from the keyword to the intent.

Two elements prevent reading the phenomenon as total substitution. First: real behavior is hybrid — users choose AI to explore and synthesize, but return to traditional search when accuracy is critical (NN/g). Second: alongside the [articulation barrier](https://www.linkedin.com/pulse/prompt-driven-ai-ux-hurts-usability-jakob-nielsen/) there is a [discoverability barrier](https://www.nngroup.com/articles/ai-information-seeking-keyword-foraging/) — the opacity of generative systems makes it hard to understand what could be delegated, and most people use slightly more verbose versions of old queries. Now we have the measure: in a study of 937 participants with analysis of 747 real conversations, **only 19.1% of users employ at least one prompting strategy**; the vast majority simply type a question ([*Scientific Reports*, March 2026](https://www.nature.com/articles/s41598-026-42465-4)). Those who do get measurably better experiences: the ability to express articulated intents remains unequally distributed, and it pays.

For content producers the consequence is direct: you need to know the most common search intents in your sector and produce content that satisfies them, because AI rewards the content that best answers the intent.

## The decision this observation informs

The decision is whether to write for intent or for editorial voice, when the two diverge. It falls under the [ATTRACT](https://www.shaped-by-ai.com/en/impacts/the-decisions-ai-forces/#attract) position. Without an explicit choice, intent wins, because citability is measurable and brand voice is not: the text flattens onto the structure AI extracts best — density, self-contained paragraphs, explicit definitions — while the citation chart goes up.

Optimising for extraction means accepting that the brand's tone and personality flatten out. Defending the voice can cost a share of citations in a channel where only 38% of citations come from the organic top 10 — where ranking no longer protects a brand on its own. The choice belongs to the Content Strategist with an explicit brand mandate, not to SEO: this remains an open node, holding together editorial voice and answering intents, precisely because it is unassigned.

## Effects and role responsibilities

* **[UX Researcher](https://www.shaped-by-ai.com/en/impacts/roles/research-ia/)** — Both the spontaneously expressed intent and the one the user does not know they can express must be mapped: due to the discoverability barrier, observed behavior underestimates what users would like to delegate. The 19.1% figure gives a starting point: designing for the non-prompter is the normal condition, not the edge case. → Useful resource: [How AI Succeeds (and Fails) to Help People Find Information — NN/g](https://www.nngroup.com/articles/ai-information-seeking-keyword-foraging/)
* **[Information Architect](https://www.shaped-by-ai.com/en/impacts/roles/research-ia/)** — Information architecture must evolve too: classic methods — card sorting, tree testing — validate a single hierarchy, but users who start from their own case skip it and expect direct, intent-based access. **Information architecture**, which organises content by domain logic and answers "where is this thing?", needs an **architecture of intent** alongside it, which starts from the user's goals and answers "how do I get X?".
  In practice it is built in four steps. **(1) Collect the real intents** from sources that already exist in the company: internal search engine queries, support tickets, call center transcripts, assistant logs if any — not from business categories. **(2) Normalize them** into a finite list of goals expressed in the user's words — "cancel the subscription", "change the IBAN", "understand why I was charged this amount" — not in internal ones ("Registry management", "Dispositive area"). **(3) Give each intent a direct access point**: a highlighted CTA, a home shortcut, a deep link, an answer at the top of the page — reachable without crossing the hierarchy. **(4) Make the two levels coexist**: the structure remains for those who explore, intent access serves those who already know what they want and are the majority of returns.
  Example: a banking site that archives direct debits under "Products › Accounts › Ancillary services › Direct debits" is not wrong — it is just unusable for someone arriving thinking "I need to set up the direct debit for my electricity bill". The architecture of intent adds that item as a first-level entry point, without dismantling the taxonomy.
  The intent map produced by this process becomes an architecture deliverable on a par with the sitemap, to be validated against the real variance in formulations — the same need produces formulations with near-zero similarity (0.081, figure cited above) — not against a closed card sort that assumes categories are already known. Governance note: intents change faster than taxonomies and must be periodically reviewed on internal search and support data, otherwise direct access ages worse than the structure. → Useful resource: [Rethinking Jobs-to-be-Done for the AI Era — Momentum Design Lab](https://assets.website-files.com/63bba9873fa686465ba767d8/654db74265088acbde8a9cc7_Cognitive%20Design%20Approach%20Momentum%20Design%20Lab.pdf)
* **[UX Writer / Copywriter](https://www.shaped-by-ai.com/en/impacts/roles/content-editorial/)** — Navigation microtexts must anticipate intent, not describe a link's destination. Text designed for AI tends toward density and predictable structure: avoid flattening the brand's tone and personality.
* **[Content Strategist](https://www.shaped-by-ai.com/en/impacts/roles/content-editorial/)** — The keyword loses centrality (AI still queries Google: many principles remain valid). What matters is answering an intent completely: informative titles, opening summaries, answers to implicit questions. The operational criterion is citability: the goal is no longer just to rank, it is to be extracted. Open node: holding together the brand's editorial voice and answering intents. → Useful resource: [AI Search Statistics 2026 — Searcherries](https://searcherries.com/ai-search-statistics/)
* **[SEO / GEO Specialist](https://www.shaped-by-ai.com/en/impacts/roles/seo-ai-visibility/)** — GEO depends on factors other than the keyword: perceived authority of sources, clear semantic structure, citability. An extractable piece of content lets the AI isolate a single claim without inferring it from context: heading hierarchy consistent with concepts, schema.org markup, self-contained paragraphs, lists, tables, explicit definitions. **The structural change of 2026 is the decoupling between ranking and citation**: only 38% of citations in AI Overviews come from pages in the organic top-10, versus the previous 76% (Ahrefs *Brand Radar*, March 2026, on 863,000 SERPs and 4 million URLs). Being first and being cited are now two distinct goals. The channel remains volatile by construction: when Google changed the default model of AI Overviews in January 2026, about 42% of cited domains were replaced (SE Ranking). New lever: **Preferred Sources**, extended to AI Overviews and AI Mode since 27 May 2026, with over 345,000 sources already selected by users and a declared double click-through probability when a preferred source appears. On the analytical side, cross multiple sources: long-tail from Search Console, Search Ads keywords, AI queries from Bing, internal engine searches.
* **[Digital PR / Reputation](https://www.shaped-by-ai.com/en/impacts/roles/brand-pr/)** — Only 1% of sources cited by LLMs come from brand-owned sites (McKinsey, 2026): the rest is earned — forums, reviews, comparisons, third-party articles. Reviews, presence in sector forums, authoritative mentions, and consistency of brand data across sources become primary GEO levers: digital PR moves from a support activity to the infrastructure of AI visibility. → Useful resource: [AI social listening — YouScan](https://youscan.io/blog/ai-social-listening/)