This week CNN ran a feature with a headline most people read and scrolled past: “AI is changing the internet forever. Here’s how.” Worth not scrolling past. Inside is the clearest signal yet that the way humans interface with the internet has structurally changed — and Google is the company most directly confirming it.
Robby Stein, VP of Product for Google Search, told CNN that people are now asking questions long enough and complex enough that “no clear response exists anywhere on the internet”. Google’s answer is to stop returning a ranked list of pages and start generating the answer itself — complete with custom visuals, interactive graphics, and mini-apps assembled live on the search page.
That is the biggest change to the search bar in 25 years. Three numbers from the same reporting make the scale of the shift concrete.
The three numbers
3 → 10 words per query. +393% YoY AI-driven traffic to US retail sites. 25 years since the search bar last changed this much. Together they say: your customers (and your employees) are now briefing a model, not searching a database.
What actually changed
It’s worth saying clearly what is new versus what is incremental, because most teams have heard “AI in search” for two years and tuned it out. What ships now is genuinely different in three ways.
1) The interface is generative, not retrieval. Old search: keyword in, ranked links out. New search: complex question in, a generated answer with custom graphics, an interactive map, a calculator, or a small purpose-built app, all rendered on the results page. The user often never clicks through to a source at all.
2) The query is a brief, not keywords. Five-to-ten-word inputs aren’t searches — they’re delegations. The shape of the input is moving toward how you’d explain a task to a competent assistant.
3) AI is the new shopfront. Traffic to US retail sites from AI services grew 393% YoY in Q1 2026. People are starting buying journeys inside an assistant, not on a brand’s homepage. For most businesses that’s a channel that didn’t exist twelve months ago.
Using AI search well versus old search is like the difference between a vending machine and a butler. With the vending machine you understood the rules — pick the row, pay, get the snack. With a butler you have to know how to brief them, when to push back on what they bring you, and when not to trust them. Most workplace AI training is still teaching people how to use the vending machine. — Lisa O'Reilly, Director, TESS Group
The skills implications most teams are missing
Five competence areas come up in every L&D conversation we’re having right now — only the first is the one most teams already have a tick-box answer for:
- Briefing. Structured delegation — context, constraints, output format, success criteria. The difference between a useless and useful AI answer.
- Output evaluation. When the engine generates a paragraph, a chart, or a mini-app, you can’t fall back on “the source ranked first, it’s probably right”.
- Source verification at speed. Generated answers cite inconsistently and sometimes invent. Fast routines matter when work goes to a customer, regulator or board.
- Workflow redesign. If AI is now doing the first-draft work in 30 seconds, the human role moves up: judgement, framing, decision.
- Governance and risk. Citing AI-generated content without verification is a Consumer Duty risk in financial services and a fitness-to-practise risk in clinical settings. We covered the regulatory frame in our UK AI compliance guide.
Want a workforce-skills plan for the AI-first internet?
25-minute discovery call. Bring your team profile and 2026 training budget. We’ll show you a three-layer plan — baseline literacy, apprenticeship depth, and leadership-unit governance.
How it maps to the apprenticeship standard
The Skills England-approved AI & Automation Practitioner Level 4 apprenticeship (ST1512) was written for exactly this paradigm. Briefing maps to AO2 (Solution Design). Output evaluation to AO3 (Testing & Iteration). Model selection to AO1 (Strategic Adoption). Audit trails to AO4 (Governance). Internal enablement to AO5 (Stakeholder & Workforce). Iteration on new tools to AO6 (Continuous Improvement).
Every implication of the CNN report sits inside the six Assessment Outcomes the standard trains. That isn’t luck — ST1512 was approved precisely because “use ChatGPT” was never going to age well as a skills definition.
Whole-team baseline literacy via short courses. Apprenticeship depth (L4 or L6) for the 5–15 people who’ll own the stack. Leadership units (AU0009/10/11) for senior decision-makers. Bring your team profile, we’ll map it.
The marketing-team take on this story will be “SEO is dead, AEO is the future”. That’s true and your marketing team should be thinking about it. But framing this as a marketing problem badly underplays it: every team that works with information is now in scope — procurement, finance, ops, customer-facing teams, HR. Each needs the briefing, evaluation, verification and governance competencies above. None of them sit in the marketing team.