The Receipts

We literally asked the engines

This is not a guess or a model of how AI might behave. We ran 12 real queries through ChatGPT and Perplexity, both with live web search, and saved exactly what came back. Here are the receipts.

The Method How the test was run

Honesty matters more than drama here, so the method is laid out in full. Anyone can reproduce this. There is nothing proprietary about the questions; they are just the ones buyers ask.

12
Real queries

Eleven category questions a buyer would type, plus one branded control ("Mads Singers") to prove the engine knows who he is.

2
Engines, web-backed

ChatGPT (gpt-4o) and Perplexity (sonar-pro), each with live web search switched on, so they answered using the current web, not stale memory.

~$0.32
Cost to run

The entire 24-answer test cost about 32 cents through the DataForSEO LLM Responses API. Real calls, captured answers, real source lists.

Each query was sent to both engines. We then read every answer and marked it Cited if "Mads Singers" or madssingers.com appeared verbatim, and Absent if it did not. No interpretation, no partial credit. Either the name is in the answer or it is not. We also kept the source domains each engine pulled from, which is where the most useful detail turned up.

Why this beats "just go ask ChatGPT yourself"

A single manual ask is anecdotal and changes run to run. Running all 12 through the API in one pass, with web search on, gives a clean, repeatable snapshot of what a real buyer would see today. It is the difference between a hunch and a measurement.

The Citation Matrix Every query, both engines, what came back

The full scoreboard. Eleven category queries returned Absent on both engines. Only the branded control returned Cited, on both. That pattern is the whole story in one table.

#Query a buyer would askChatGPTPerplexity
1best people management coachAbsentAbsent
2DISC assessment training expertAbsentAbsent
3how to delegate effectively as a business ownerAbsentAbsent
4management coach for entrepreneursAbsentAbsent
5how to get employees to take ownershipAbsentAbsent
6team management consultant for scaling agenciesAbsentAbsent
7best management course for entrepreneursAbsentAbsent
8how to run effective 1:1 meetings with your teamAbsentAbsent
9people management mastery courseAbsentAbsent
10how to hire the right peopleAbsentAbsent
11executive coaching for delegationAbsentAbsent
12Mads Singers (branded control)CitedCited
Read down the two engine columns: they agree on every single row. Eleven Absent, one Cited, identical on both. When two independent engines, using different models and different web indexes, return the exact same pattern, that is not noise. That is a stable, real-world result.

Who Got Cited Instead The roster the engines reached for

For each of those 11 category questions the engines did name someone, just not Mads. Here is exactly who filled the slot, grouped by the kind of source.

The author roster

The default "management expert" names the models lean on:

  • Marshall Goldsmith
  • Simon Sinek
  • Brene Brown
  • Liz Wiseman
  • John C. Maxwell
DISC vendors

For the assessment and DISC queries, the engines named tools, not people:

  • Thomas International
  • Everything DISC
  • Discflow
  • DISCGB
Platforms

And the marketplaces and big-brand platforms that aggregate coaches:

  • MentorCruise
  • Tony Robbins
  • FutureLearn

Notice the shape of that roster. Several are authors who do not run one to one coaching at all, several are software vendors, and several are platforms that take a cut. None of them out-coach Mads. They win the slot because they are the names and brands the engines have the most structured, repeated, cited signal about. Recommendation eligibility is what they share, not coaching quality.

The source-domain tell

When we looked at where the engines pulled their answers from, one domain dominated: YouTube appeared roughly seven times across the source lists, far more than any other, alongside a cluster of UK coaching and training sites. The encouraging read: YouTube is already a place Mads publishes. The engines clearly trust it as a source. The work is making his YouTube and site presence the version the engine reaches for, rather than someone else's.

Coverage What we tested and what we did not

A quick honest note on scope, so you know exactly what is behind the headline and what is not, with no rounding up.

Tested in full

ChatGPT and Perplexity

Both engines probed across all 12 queries, web search on, answers and source lists captured. This is the complete dataset behind the matrix above.

Deliberately skipped

Google AI Mode as a third engine

We did not add a third AI engine. The two web-backed engines already agreed perfectly at 0 out of 11, so a third would have spent budget to confirm a result that was already unanimous. We would rather be honest about that than pad the number.

Page performance and the technical render (Lighthouse, schema, sitemap, stack) were captured separately and live on 07 - SEO + Tech. The strategic read of this whole test, including why a surfacing gap is the best kind of problem to find, is the headline on 05 - AI Visibility.

Why show this at all. Mads coaches Donal on the business, so turning the agency's measurement machine on his own site is a small thank you and a way to show what it actually does. The result is genuinely hopeful: the engines already know who he is, they just have not been given a reason to recommend him yet. If any of this is useful, take it and run. If not, no pressure at all.