The Agent Promoter Score standard
The Agent Promoter Score (APS) measures how AI agents describe a company to the people who ask about it. This page defines the score, its four pillars and the rules for reporting it, so every company is measured the same way.
| Identifier | APS-1.0 |
| Scale | 0 to 100 |
| Pillars | Four defined, three scored |
| Cadence | Weekly, reported as a 12-week trend |
| Method | APS-1.0 methodology |
| Publisher | The Prompt Group |
Where APS fits
Every choice a buyer makes with an AI agent passes through the same four checkpoints. Most of the market measures the second: whether a company appears at all. APS measures the third: how the company is described once it does.
| Checkpoint | The question it answers | What is measured |
|---|---|---|
| 1. Model selection | Which AI does the buyer ask? | Each engine draws on different sources. |
| 2. Inclusion | Am I on the shelf? | Whether a company is named, and in what order. |
| 3. Agent Promoter Score | How am I described? | How accurately, favourably and loyally the agent speaks about the company. |
| 4. Choice survivorship | Do I hold up? | Whether the company survives the follow-up questions. |
The four pillars
Like the Net Promoter Score it is named after, APS is a loyalty measure. Here, the one being loyal is the AI agent. Each pillar asks a different kind of question and is scored from 0 to 100.
Pillar I: Knowledge accuracy and factuality Scored
Are the facts the agent states about the company correct? Prices, products, terms and people.
What's the price of [company]?
Pillar II: Subjective perceived value Scored
When a buyer asks for an opinion, how favourably does the agent speak about the company?
Is [company] a good choice for me?
Pillar III: Competitive loyalty Scored
When buyers ask for options, how often is the company named, and how near the top?
Best [category] for my team?
Pillar IV: Strategic outcome effectiveness In development
Does the answer move the buyer to the next step, such as a demo or a booking? The method goes to consultation before it enters the score.
How do I book a demo with [company]?
Computing the score
Under APS-1.0, the score is the weighted sum of pillars I to III. Because each pillar runs from 0 to 100 and the weights sum to one, the score does too.
// APS-1.0 composite APS = wI·I + wII·II + wIII·III // provisional weights, until weight testing is complete wI = wII = wIII = ⅓
- Weights are published with every version, so scores from the same version compare directly.
- A score is published only when every scored pillar has its minimum evidence. See The composite score.
Case studies
Both companies below were recommended by the agent, and both lost the sale. Neither loss would show up on a visibility report.
The airline that lost on a currency label Pillar I
An administrator books flights for a company party. The agent recommends a carrier, then quotes the fare in US dollars without saying so. Converted to Canadian dollars, the trip is over budget, and she moves on.
Failed on knowledge accuracy: a fact was stated wrongly.
The SUV that lost a question it never answered Pillar II
A parent with two car seats and a long winter commute asks whether a shortlisted SUV suits her. The agent calls it a city crossover and points her to two other names.
Failed on perceived value: there was no story for the agent to find, so it wrote one.
Specimen report
Every APS report goes down to the question, so a team can see which answers cost them points, and why.
| Question | Pillar | Score |
|---|---|---|
| What does a return flight to Vancouver cost? Fare quoted without currency in 4 of 6 engines. | I | 38 |
| Is this airline good for a group booking? Described as reliable; group desk not mentioned. | II | 71 |
| Should I fly with them or a competitor? Recommended first in 3 of 6 engines. | III | 68 |
| How do I book for 40 people? Collected, not scored under APS-1.0. | IV | — |
Further reading
Where the four checkpoints and the APS pillars were first defined.
Read the paper ↗ BenchmarkThe Answer IndexWhich companies AI engines name, category by category.
See the index ↗ MethodAPS-1.0 methodologyEvery step from a buyer's question to the number on the report.
Read the method →Citing APS
APS is free to use and cite. When you report a score, report it with:
- the version that produced it, such as
APS-1.0 - each pillar score, and any pillar not scored
- the engines measured, the number of answers and the collection window
Citing the standard
The Prompt Group (2026). Agent Promoter Score Standard, Version 1.0 (APS-1.0). Draft for public consultation. agentpromoterscore.com
Reporting a score
APS 64/100 · APS-1.0 · I 52 · II 71 · III 68 · IV not scored 6 engines · 1,840 answers · Jul–Sep 2026