Research

The evidence behind recognition.

The Recognition Triangle is a model. This page is the evidence for it: independent, primary research on what actually moves an AI system to cite, trust, and recommend a brand, mapped to the three conditions, Clarity, Credibility, and Earned Authority. Every figure below names its source, its date, and what it does not prove.

In one paragraph

This page collects the independent research that supports the Recognition Triangle: peer-reviewed findings on what actually moves AI citations, large-sample industry indices on who gets cited and how often, and national survey data on how many people now ask AI instead of searching. Every figure is checked against its primary source and shown with its date. Aingworth's own Recognition Baseline Study, our first primary dataset, is described further down this page and will be added here in full once the first run is complete.

The most persuasive claim on this page is the one that shows its work.

Our Claims Standard says every statistic we publish carries its source inline and is treated as a hypothesis to test rather than a settled fact. This page is that standard applied in one place.

It exists because the shift underneath all of it is no longer a guess. Search became answer-driven, most of what a machine says about a brand lives on sites that brand does not own, and almost half of marketing leaders admit they cannot measure any of it properly. The sections below show that evidence mapped to the three conditions of recognition, then explain the study we are running on ourselves to add first-party evidence of our own.

Why Measure At All

The audience is real. The measurement mostly is not.

49%

of US adults now use an AI chatbot, up from 33% in 2024, and 24% use one daily - Pew Research Center, June 2026

45%

of marketing leaders cannot accurately measure their own brand's AI visibility; only 9% have tooling that actually tracks it - Semrush, 2026

36

global brands hold top-100 visibility across ChatGPT, Gemini, and Google's AI surfaces every single month, out of 126 million prompts analyzed - Semrush, 2026

Mapped to the Triangle

Three conditions. The evidence for each.

Recognition forms only where Clarity, Credibility, and Earned Authority all hold at once. Here is what the independent research says each one actually depends on.

Condition One

Clarity

  • Fluency and clean structure were among the largest tested gains in the founding academic study of this field, adding up to 15–30% to a source's visibility in generative answers, while keyword stuffing, the classic SEO move, showed minimal or negative returns - GEO, KDD '24, Princeton/IIT Delhi, 2024
  • Different engines are structurally different reading environments: ChatGPT cites an average of 15 sources per response, Gemini cites only 3, which means the same brand needs a differently shaped Clarity story depending on which machine is asking - Semrush, 2026

Condition Two

Credibility

  • Adding statistics to a page lifted visibility by 30–40%, and citing outside sources was the single largest tested effect of all, a 115% gain, specifically for lower-ranked sources that had little authority to begin with - GEO, KDD '24, 2024
  • A controlled study of 1,885 pages against 4,000 matched controls found schema markup produced no statistically meaningful citation gain on ChatGPT or Google AI Mode. Credibility is substance a system can verify, not a tag it can read - Ahrefs, 2026

Condition Three

Earned Authority

  • Wikipedia and Reddit alone now account for more than a quarter of ChatGPT's US citations, and LinkedIn jumped from the 11th most-cited source to the 5th in three months - 5W, 2026
  • Reddit's own citation share collapsed from roughly 60% to 10% within two weeks in September 2025, which is the strongest argument we have found that Earned Authority has to be monitored continuously, not audited once and filed away - 5W, 2026
Deep Dives

One framework at a time, in full.

Each page below takes a single Aingworth framework and shows the independent data behind it in full, charts included, not just a link out. We are building these one at a time, starting with the mechanism that decides who gets shortlisted before a single word is ever written back.

Framework 01 · Live

The Machine Gatekeeper

What actually correlates with getting cited across 129,000 domains, Google's own words on optimizing for AI, and why most AI crawlers get blocked before they ever reach the page.

Read the evidence

Framework 02 · Live

Earned Authority

The Structural, Expert, and Community layers, measured across 680 million citations, why fifty domains decide most of what AI says, and how fast that can change.

Read the evidence

Framework 03 · Live

The Recognition Economy

Why one engine's dominance fell by a third in eight months, how Google itself is rebuilding for named answers, and why concentration isn't the same in every category.

Read the evidence
In Progress · Aingworth's Own Data

Our first primary study is running now.

Every source above is someone else's measurement. This one will be ours. The Recognition Baseline Study runs our ten priority buyer questions, the same questions a prospective client would actually type into an AI system, thirty times each, per model, using the Recognition Check, our own open-source instrument. For each run we record whether Aingworth, Lucy Aingworth, or Become the Answer is named, which other names come up instead, and, in search mode, which sources get cited.

Thirty runs per prompt is not an arbitrary number. A single AI response is not a stable measurement: ask the same question twice and you can get two different answers, so a handful of runs is not enough to see a real pattern rather than noise. Thirty is the point where that pattern settles down enough to trust, and it is the same standard we hold every client measurement to.

This is also case study number one: Aingworth measuring itself, in public, with the same instrument we sell. We would rather publish an honest first number and watch it move than wait for a flattering one. The full write-up, method and limitations stated plainly, lands on this page as soon as the first run is complete, and reruns monthly after that so the number stays current rather than frozen at a single, convenient moment.

METHOD · 10 buyer questions × 30 runs × model
RECORDS · named y/n · competitors named · sources cited
OUTPUT · Recognition Rate per prompt, and overall
CADENCE · rerun monthly, published as it moves

Run the Recognition Check Read the Recognition Triangle
Common Questions

Questions people ask about this research.

Is this Aingworth's own research?
Mostly not, and we say so on every line. Almost everything on this page is independent, primary research published by universities, industry analysts, or national research bodies, cited with its own source and date. Our own first-party dataset, the Recognition Baseline Study, is separate and clearly marked as in progress.
What is the Recognition Baseline Study?
Aingworth's first primary study: ten real buyer questions, run thirty times each per AI model, using our own open-source Recognition Check instrument, measuring how often Aingworth is named and which sources get cited. It is dated, method-disclosed, and rerun monthly so the number can move over time rather than sit frozen as a one-off claim.
Why thirty runs per question, not five or ten?
Because AI outputs vary run to run. Ask the same question a handful of times and the answers can differ enough to be noise rather than a pattern. Thirty runs is the point where that pattern settles down enough to trust, and it is the same standard we hold every client measurement to.
How often is this page updated?
As new independent research is published and verified at the primary source, and every month as the Recognition Baseline Study reruns. Given how fast citation patterns move, a study older than a year is treated here as historical context, not a current claim.

A note on verification

Figures on this page are checked against the primary source wherever one is publicly available; where only a press release or a secondary write-up could be reached, that is the link provided, and it is treated as a lead worth reverifying rather than a settled fact. We do not claim to have caused any pattern we only observed. If you find a figure here that has been superseded or misquoted, tell us at [email protected] and we will correct it.

See where your own brand stands.

Run the free diagnostic to find which condition of the Triangle is leaking, or run the Recognition Check to see how often AI actually names you today.

Run the Free Diagnostic Run the Recognition Check