By Riddhesh Patil6 min read

Google Ranks You #1. ChatGPT Has Never Heard of You.

Picture a buyer with a real problem and a budget. Ten years ago they typed a few words into Google, opened four tabs and started comparing.

Today many of them open ChatGPT, Gemini, Perplexity or Copilot and ask a full question: "What's the best AI automation agency for a midsize UK firm?"

They get back a confident paragraph that names three companies, says a line about each, and sometimes cites a source or two.

If your brand isn't one of those three names, you weren't outranked. You were left out of the conversation entirely.

A search for the best AI automation agency for a mid-size UK firm, first as ten blue links to three competitors, then as an AI answer that names the same three agencies
[ fig. 01 ]From ranking in the links, to being the answer in the box, to being the brand the AI names.

Three eras of being found

SEO (search engine optimisation) was about the ten blue links. Rank higher, get the click. The page did the persuading.

AEO (answer engine optimisation) arrived with featured snippets, People Also Ask and voice assistants. The goal shifted from ranking to being the answer: one source quoted directly at the top of the page.

GEO (generative engine optimisation) is where we are now. AI engines don't quote one page. They read many, combine them, and write their own answer. Sometimes they name brands and sometimes they don't. They cite sources, but not always yours.

These eras don't replace each other, they stack. Many AI answers are built from search results: Copilot draws on Bing, and Google's AI Overviews draw on Google. That makes good SEO the foundation, and often a better ranking is the first sign of a better AI result. But ranking alone no longer guarantees you get mentioned.

An AI answer to the same question lists Competitor A, B and C with a reason for each, while the Squawk duck holds a sign reading "You: not mentioned"
[ fig. 02 ]The new shelf space: three names in an AI answer. Is yours one of them?

The problem: everyone's guessing

Most "AI visibility" advice today rests on a screenshot. Someone asks ChatGPT a question once, sees their brand (or doesn't), and draws a conclusion.

That approach has three problems:

  1. AI answers vary. Ask the same question five times and you can get five different shortlists.
  2. Engines change constantly. Models and search back ends update without notice, so this week's answer isn't comparable to last month's.
  3. No baseline means no proof. If you can't say where you started, you can't prove that anything you did made a difference.

At Pilotree we kept hitting the same wall. Clients asked a simple question, "Are we showing up in AI, and is it getting better?", and the honest answer was "we can't tell yet." So we built a way to tell.

How we measure it: Squawk

Squawk is our internal search and AI visibility platform. It measures how visible a brand is in Google and in AI answers, plans and logs the work that improves both, and proves the result with numbers a client can check for themselves.

The method follows a few rules.

1. Ask like a buyer, and ask more than once

Every client gets a set of 20 to 40 questions, written the way a real buyer would phrase them and grouped into topics. They fall into four kinds:

  • Brand: "What does [brand] do?" Does AI describe you correctly?
  • Category: "Best [service] for [buyer] in [place]". Are you named when the buyer doesn't know you yet? These are the questions that win new business.
  • Comparison: "[Brand] vs [competitor]". How are you positioned against rivals?
  • Control: a real question on a topic we will not work on. This is the yardstick (more on it below).

Each question is asked five times on every engine: a ChatGPT stand in, Google AI Overviews, Gemini, Perplexity, Copilot and Claude. Every answer is stored in full, with its date, model version and citations, so every number can be traced back to a stored answer.

A grid of one question asked five times on ChatGPT, AI Overviews, Gemini, Perplexity, Copilot and Claude, filling in as answers name the brand, while the Squawk Score rises from 10 to 37
[ fig. 03 ]One question, five samples, six engines. The Squawk Score is the share of those answers that name you. (Illustrative data.)

2. One headline number

The Squawk Score (0 to 100) is the weighted share of AI answers that mention you, counted on category and comparison questions only. Brand questions always name you, so they're left out. Control questions are the yardstick, so they're left out too.

A mention only counts if it's really you. If an AI talks about a different company with a similar name, that's logged as a confusion, not a mention. Confusions are a useful finding in their own right.

3. Separate real change from noise

This is the part most AI visibility reports skip.

Each metric is reported with a noise band, a 95% range. A change smaller than that range isn't claimed as a win.

The control questions then catch the engines changing under you. Say your category questions rise 12 points and the control questions, which nobody touched, rise 5. The engines moved by roughly 5 points on their own, so the lift you can actually claim is 7.

Two bars: category questions rise 12 points and control questions rise 5, leaving 12 minus 5 equals 7 points of lift that can be claimed
[ fig. 04 ]Engines drift. Control questions measure that drift, so it isn't counted as your win.

When the control moves beyond its own noise band, the report says so plainly and holds off on claiming a win.

4. Find out who AI cites instead of you

For every topic, Squawk lists the domains the engines cite and sorts them into six types: own, competitor, editorial, community, reference and directory. The mix decides what to do next.

  • Competitor pages are winning? Build or improve your own pages on that topic.
  • Editorial or community sources are winning? Earn coverage there. Another page on your own site won't help.
Cited domains sorted into six columns, own, competitor, editorial, community, reference and directory, ending with the advice that when editorial and community sources win, you should earn coverage there
[ fig. 05 ]Follow the sources. The citation mix tells you whether to build pages or earn coverage.

What actually moves the needle

The good news is that most fixes improve SEO and GEO at the same time:

  1. Deeper pages on the topics buyers actually ask about. Thin service pages lose to competitors who answer the full question.
  2. A named, credible author or expert on the content.
  3. Structured data that makes it easy for machines to read who you are, what you offer and where.
  4. Consistent facts across the web. If your site, directories and press coverage disagree about what you do, AI repeats the confusion.
  5. Presence where AI looks: industry press, community threads and the directories that keep showing up in citations.

None of this is new. What's new is being able to show that it worked.

Intelligent growth means measured growth

The future of digital marketing isn't SEO or AEO or GEO. It's all three, measured together, on the same topics, with a fixed method and a clear before and after.

Rankings still matter. Being the answer still matters. But the question that decides the next decade of discovery is simpler:

When your buyer asks AI, does it say your name?

If you don't know the answer, that's the place to start.

Squawk helps companies get found in Google and in AI answers, and proves it with numbers they can check. Want to know whether AI names your brand? Get a free visibility check.