A WORKING VIEW

How do people find and choose a company now?

People find companies in places most reports can’t see: AI answers, social feeds, private messages, and what friends say. The click you track is often the last step of a decision made somewhere else. I think the work is to show up where people form their opinions, and to measure with methods that don’t depend on a click.

Updated November 17, 2026. I revise this page as the evidence changes.

The map

Budgets sit where you can control and measure. Much of discovery may happen where you can’t.

← Less controlMore control →
← Hard to measureEasy to measure →

Where each channel sits is my judgment, not a measurement.

My current view, in six points

Where are people deciding about us, and would we know if we were missing it?

WHAT CHANGED

Fewer decisions end in a click you can see.

Search still matters. But more searches end on the results page, AI summaries answer before anyone clicks, and more people take in information from feeds.

ZERO-CLICK SEARCH

58.5%

Of U.S. Google searches ended with no click.

SparkToro and Datos, data from Sept 2022 to May 2024

AI SUMMARIES

8% vs 15%

Visits where people clicked a regular result, with an AI summary on the page and without one.

Pew Research Center, March 2025 data

NEWS IN FEEDS

1 in 5

U.S. adults regularly get news on TikTok, up from 3% in 2020.

Pew Research Center, August 2026

People still click. But by then most of the decision is made.

01

Most of the decision happens before the click

The question to ask

Where did this buyer first hear about us?

Ask buyers where they first heard of you, and you often get a different answer than your software gives.

Same buyers, two answers

Software credited search. Buyers named social, podcasts, and word of mouth.

Web search

Software said78%
Buyers said12%

Social, podcasts, communities, word of mouth

Software saidNot tracked at all
Buyers said85%

Share of 620 conversions over 12 months at one company, Refine Labs. Buyers could name more than one source. Refine Labs sells this approach, so treat it as a reason to run the test yourself.

Click data shows the same thing. Most searches end with no click, and people click about half as often when an AI summary shows. (Search Engine Land, Pew, Refine Labs)

If you get this wrongYou credit the last step and cut the ones that started it.

02

Each way of measuring sees a different part

The question to ask

Which question is this report actually answering?

No single method sees the whole picture. The mistake is using one to answer another’s question.

MethodWhat it seesUse it to answer
Click trackingThe last tracked step before a saleWhich ads and pages close the deal
Asking buyersWhere they remember first hearing of youWhere demand starts
Holdout testsWhat a channel adds that wouldn’t have happened anywayIs this spend causing results
Mix modelsEach channel’s share of results over timeHow to split the budget
Branded searchWhether more people look you up by nameWhether people know you better

Tests matter because tracking credits ads for sales that would have happened anyway. When eBay switched off ads on its own brand name, it found no measurable short-term benefit. (NBER) In 15 randomized tests at Facebook, the usual statistical methods often missed what the tests found. (Kellogg) Google’s free Meridian makes mix models easier to start, but a model from the company selling the ads still needs an outside check.

If you get this wrongOne report settles every question, including the ones it can’t answer.

03

Organic social works through memory

The question to ask

Are more people looking us up by name?

Organic social puts you in front of people before they’re looking. Most of what it does won’t show up as a click.

Seen first, searched later

In controlled tests on Yahoo, display ads from three advertisers raised searches for the advertised brand by 30 to 45%.

A Harvard Business School study found the lift in searches showed up after about two weeks, not right away.

Lewis and Nguyen, Quantitative Marketing and Economics, 2015. Harvard Business School researchers, via Search Engine Land. These studies measured paid display ads, not organic posts. I haven’t found a test like this for organic social.

Feeds are where many people take in information now. 1 in 5 U.S. adults regularly get news on TikTok. (Pew) And links shared by text or Slack usually arrive with no source. So check social with branded search and by asking buyers, not with referral traffic.

If you get this wrongYou judge it by clicks and cut it.

04

Thought leadership reaches people and AI

The question to ask

What do we know that others would want to quote?

Thought leadership shapes what buyers think before they talk to you, and it gives AI tools something to cite.

In the market right now

About 5% of B2B buyers are in the market at any one time. The other 95% buy later, so most marketing works by being remembered.

John Dawes, Ehrenberg-Bass Institute

People in the buying group who never talk to sales read thought leadership much like the main buyers do. (Edelman and LinkedIn) AI answers lean on a few sources, and the list changes fast. In September 2025, ChatGPT’s Reddit citations fell from close to 60% of responses to about 10% within days. (Semrush, Ehrenberg-Bass)

What lasts: a specific point of view, under a person’s name, published on a schedule, with original data others have to cite.

If you get this wrongYou publish more often, but nobody can tell what you think.

05

Cutting one channel weakens another

The question to ask

If we cut this, what else gets weaker?

Each channel does a different job. Cut one, and another gets weaker later, often after the report has counted the cut as savings.

The chain

Reports credit the last step. The earlier steps are what made it possible.

  1. Step 1Hear about youSocial, thought leadership, creators, word of mouth, AI answers to general questions
  2. Step 2Remember youSteady posting, PR, reach-based ads
  3. Step 3Look you upSearch, your site, AI answers about you
  4. Step 4Choose youPaid search, retargeting, sales
Hard to measureEasy to measure

AI answers show up twice. They can introduce you to someone who has never heard of you, and they describe you to someone looking you up.

  • AI answers introduce companies to people who aren’t looking for them yet. In a survey of 1,030 U.S. shoppers who use AI tools, 43% had found a new brand through one. (Semrush) Among U.S. business buyers who use AI at work, 63% search for a company on Google after AI mentions it. (Semrush) So an AI answer can start the search that your report later gets credit for. Semrush sells AI visibility tools, so treat these as a reason to ask your own buyers.
  • The names in those answers come from what’s published about each company on other sites. Cut the work that gets you mentioned, and you can drop out of those answers. (Semrush)
  • Across 996 award-entry campaigns, those that combined brand building and short-term activation did best. Binet and Field suggest roughly 60% brand, 40% activation. (IPA)
  • Airbnb cut marketing heavily in early 2021, mostly performance spend, and said traffic held near 2019 levels. It was already a household name, so most companies can’t copy that. (Marketing Week)

If you get this wrongA cut shows up as savings, and the loss shows up later in another channel.

06

Set the money rules first

The question to ask

What would we need to see to shift budget?

Split the budget in two, judge each part by a different standard, and decide in advance what would make you change it.

Two budgets, two standards

Judge closing spend by tests. Judge creating spend by early signs.

Spend that closes demand

Paid search, retargeting, offers

Fund it by
What holdout tests show it adds, not what tracking credits it with
Test first
Ads on your own brand name

Spend that creates demand

Social, thought leadership, PR, reach-based ads

Fund it by
A set share of the budget
Judge it by
Branded search, direct visits, buyer mentions, AI mentions
Write the rule firstExample: if branded search and buyer mentions don’t rise in two quarters, cut it by a third.
One holdout test a quarterOn the channel tracking credits most. Four real answers a year.

If you get this wrongThe channel with the cleanest dashboard gets the budget.

Still open

Where I could be wrong

These are the points where the evidence is weakest, or most likely to change. I’ll update this page when it does.

01We don’t know how much discovery we can’t seeThe strongest numbers come from companies that sell social, podcasts, or attribution tools. The click studies are solid.
02Asking buyers has its own biasPeople name what they remember, so memorable channels may get too much credit, the same way tracked ones do in software.
03Business buyers aren’t everyoneMuch of this research is about B2B. Consumers and students may decide differently, and that needs its own evidence.
04AI answers change fastWhich sources they cite shifted within days in 2025. Anything here about AI citations could be out of date in a few months.
05An AI mention may not decide muchIn one survey of 1,000 U.S. consumers, 2% would buy from an unfamiliar brand on an AI recommendation alone. 98% check somewhere else first. If that holds, search, reviews, and reputation still decide, and AI mostly gets you on the list. (Idea Grove, via Demand Gen Report. Idea Grove is a PR agency.)

NOTES

What I’ve written on this

SOURCES

SUBSCRIBE

Get new writing by email.

Essays and notes on work, growth, and whatever else I’m thinking through. You get one email when I publish something new.