Free for Australian agenciesThe first issue is in production
The Answer Reportby Aptigram Studio Get your free report
The thesis

From pages to answers

Why AI changes how buyers decide, and who can afford to understand them.

Buying is answering a chain of questions. For twenty years the buyer did the answering: search, open the pages, compare, decide. AI tools now do that work and hand back one answer. The same shift has reached the other side of the market. Understanding customers used to cost a consultancy's fee and a quarter of a year. It now costs a fraction of that, and can cover a whole market every quarter.

1How buyers decided

A buyer with a question typed a few keywords, got ten links and did the rest: read, compare, assemble an answer, decide what to do next. Marketing followed. Brands published pages that matched the keywords, and agencies built a practice on getting those pages to rank.

The buyer's own time was the engine of the whole journey. Every page was an input they had to process themselves.

2How they decide now

An AI tool takes the question whole. It splits it into smaller questions, searches for each, reads the sources, picks a few, and writes one answer, often naming two or three firms. Google now does the same at the top of its results.

In a randomised field experiment run in early 2026, an AI Overview cut clicks through to websites by 38%, and the share of searches ending without a click rose from 54% to 72%.

Illustrative. The website hasn't disappeared; its job has changed. Buyers now arrive having been told what to think, and use the site to check it.

3How firms understood customers

On the other side of the market, understanding customers well enough to design for them was a luxury. A proper piece of customer strategy covers research, personas, journey maps and prototypes. From a large consultancy it typically costs $150,000 to $300,000 and takes 12 to 20 weeks. It rests on 15 to 30 interviews and describes one organisation's customers, once.

So most businesses never buy it. The cost was labour: recruiting and interviewing, coding what people said, synthesising it, and prototyping what it suggested.

Where the money goes

Each cost line's share of a typical engagement, as a range. Our estimate, from scoping this work.

4How they can now

  1. Coding at scale. Language models code thousands of posts, reviews and searches against a codebook, with a person double-checking a sample. The most expensive hidden step becomes the cheapest.
  2. Collection by agents. Forums, reviews, search demand and every firm's website are gathered on a schedule, with a note on each source's bias.
  3. A different kind of evidence. Instead of a few dozen people in a room, thousands of people asking at the moment they had the problem, in their own words.
  4. Building collapsed. The calculator or readiness check that only existed inside a six-figure website build is now days of work.
  5. Continuity. The project becomes an operation that runs every quarter, with a person at each gate.

The result is a new capability, not a cheaper version of an old one. A single-client study can't say what the rest of a market offers. An instrument that covers the whole market does it by default.

5Four kinds of answer

Every answer used to cost someone's time. AI makes each kind of time cheap, and that is how it disrupts. Value moves to the AI platform, unless your answer needs something only you have.

0 · BeforePages

An article the buyer reads and applies themselves.

Frees nobody. Value goes to whoever ranks.
1AI answers

AI reads public sources and writes the answer.

Frees the buyer's research. Value goes to the platform, and whoever it names.
2Tools

A small web app on the buyer's own numbers. AI built it; none runs inside it.

Frees the developer's time. Value goes to you, if it needs your prices, rules or data.
3AI apps

A tool that reads the buyer's material and explains it.

Frees the expert's first conversation. Value goes to you, if it holds your knowledge.
4Agents

Something that does the job for the buyer, with permission.

Frees the doing. Value goes to whoever the buyer trusts to act.

The substitution test

A tool isn't safe because it's a tool. A generic mortgage calculator does what a chatbot already does for free. A bank's calculator survives because it runs on the bank's live rates and leads into the bank's application. Tools, AI apps and agents keep their value only when they need your data, rules, prices, commitments or systems.

Where it runs

A tool can live on your site, or inside the buyer's AI assistant as an app that runs in the chat. There it becomes part of the AI answer itself. That's the reply to "AI took my traffic": put the tool where the question is asked. It's new, and still to be proven.

One design rule for tools and AI apps: use inference at the edges and keep the core deterministic. Let the model read messy inputs and explain the result. Keep the numbers and rules fixed, so they can be reproduced and signed off.

Try it: what kind of answer does a question need?

Pick a question, or tick what's true of yours
Kind of answer AI answer

AI will answer this from public sources. Your job is to be named.

6What follows

  1. Content becomes answers. The unit is the answer unit: a claim, its proof and a point of view competitors don't share, written for a machine to use. Some answers are tools.
  2. Measurement becomes share of answer. Rank and traffic give way to how often AI names you, reported as a range because answers vary.
  3. Strategy becomes questions. Keywords give way to intents, and to the questions they break into: a question map. It models what the machine does each time someone asks.
  4. What survives is what only you have: your own data, judgement, named experts and third-party proof. Generic articles are what AI writes for free.
  5. Buyers still think, about different things. They frame the question and decide whom to trust. Trust stays human, and so does proof.
  6. One language for the whole journey. Customers keep asking after they buy: how do I set this up, what else can it do, why am I still paying? Every touchpoint answers a question. Inbound is being in the answer when they ask; outbound is bringing it before they do.

7The instrument

Put the two shifts together and a new kind of report becomes possible. Each quarter, map a market's questions: what buyers ask, what AI tells them, which tools exist. Then place every brand in it, at almost no extra cost per brand. That's the Answer Report. It runs as a loop, with a person at each gate.

ResearchWhat buyers ask, what AI says, which tools exist
MapThe market's question mapA person checks the coding
ReportEach brand's place in itA person checks each report
PlanAbout ten moves, each with its format
BuildAnswers, tools, AI appsA person signs off each build
MeasureShare of answer, tool use, enquiries

What gets built and measured feeds the next quarter's map, so the research is weighted by what worked.

8Caveats

  1. Most of the shift is still inside Google. AI Overviews are the big change. Standalone AI assistants were under 2% of US desktop search activity in early 2026.
  2. AI answers can be wrong, and they vary from run to run. Checking what AI says about you is part of the work.
  3. No study yet shows that share of answer drives revenue. We report it beside enquiries, and we'll stop leading with it if the two don't move together.
  4. Public evidence has limits. Forums over-represent the articulate and reviews the finished. Nothing represents the silent, and nothing behind a login is visible. Work that needs private access isn't replaced: a firm's own customers, its internal data, getting a leadership team to agree. The machine does the research; a person edits it at the gates.

9How we'll know

A thesis that can't be wrong is a brochure. This one is wrong if:

  • share of answer rises across quarterly reports while enquiries don't;
  • agencies who receive their report don't act on it: fewer than 2% of the first hundred reply;
  • a person checking the coding disagrees with it too often, with agreement below 0.7 (Cohen's kappa) on the core measures.

ASources

  1. Agarwal, S. and Sen, A. (2026). Randomised field experiment on AI Overviews, 1,065 US users, January–February 2026. SSRN working paper, not yet peer-reviewed. Reported by Search Engine Journal, 27 April 2026.
  2. Datos and SparkToro (2026). State of Search, Q1 2026: desktop clickstream, US, EU and UK. Reported by PPC Land, 27 April 2026.
  3. Agency Management Institute and Audience Audit (2026). Agency Core 2026, executive summary, May 2026.
  4. Basis (2026). 2026 Advertising Agency Report, April 2026.
  5. Engagement costs, timings and cost shares: our estimates, from scoping customer strategy work.
See the thesis applied to your agency

The Answer Report maps what businesses ask when they choose an agency, what AI tells them, and where you stand. It's free for Australian agencies.

Get your free report