# AgentDiscoverability.com

> AgentDiscoverability.com measures and improves whether AI agents find, shortlist, and rank a product's callable tools for the customer intents that product can genuinely fulfil.

Agent Tool Optimisation (ATO), coined by agentdiscoverability.com, is the practice of measuring and improving whether a product's callable tools are Found, Picked, and Positioned by an AI agent. It operates on the tool-selection layer: the decision an agent makes before a connection or tool call reaches the product.

Answer Engine Optimisation (AEO) wins the citation. Agent Tool Optimisation (ATO) wins the call.

## The problem we make visible

When a customer gives ChatGPT, Claude, or another agent an action Prompt—book a hotel, update a project, analyse a file—the agent decides how to complete the work. It may answer from model knowledge, search the web, operate an interface, or choose a callable tool through a ChatGPT Plugin, Claude Connector or MCP server.

If the tool route is used, the agent may search for relevant capabilities, create a shortlist, and order the options it presents. This happens before the winning product receives a connection or tool call. Products that are skipped often receive no analytics event and cannot see which competitors appeared instead.

AgentDiscoverability.com instruments this pre-connection decision. We run stable sets of commercially important Prompts repeatedly inside real agent clients, record the outcome at each selection stage, and compare the product with eligible alternatives in the same Category and Platform context.

## The three contests

1. Found — when the agent searches for capabilities that match the customer Intent, does the Integration appear?
2. Picked — after considering the full request, does the agent keep the Integration in the shortlist presented to the customer?
3. Positioned — when the Integration is Picked, where does it appear relative to the alternatives?

Found is visibility. Picked is fit. Positioned is relative priority. A product can pass one contest and fail the next.

The Discoverability Score combines repeated Found, Picked, and Positioned outcomes into a 0–100 measurement for a defined Prompt set, Category, Platform, context, and time window. It is a diagnosis of observed agent behaviour, not a universal grade or a guarantee of future selection.

Connection, authentication, invocation, and downstream task completion are important adjacent stages. The current ATO measurement product is focused on the pre-connection decision that determines which Integration gets that opportunity.

## How the product works

- Measurement Engine — shows where an Integration is Found, Picked, and Positioned across the customer Intents that matter.
- Optimisation Intelligence — identifies which contest the Integration is losing and prioritises accurate changes to its agent-facing surface.
- Studio — works with an operator from the AgentDiscoverability.com team to improve that surface, coordinate the release, and verify whether the change produced lift.

The operating loop is:

1. Measure — define a fixed Intent and Prompt set, then establish a repeated-run baseline.
2. Diagnose — locate the stage where the Integration drops out and inspect the evidence and competing surfaces.
3. Improve — change the highest-impact agent-facing metadata, tool names, descriptions, schemas, documentation, or capability boundaries.
4. Verify — rerun the same Prompts under comparable conditions and measure the movement against the baseline.

A single favourable Run is not proof. Look for sustained movement across repeated Runs and matched measurement windows.

## If you do not have a callable surface yet

AI App Studio is the build path for teams that do not yet have an Integration agents can call. We work with companies to scope valuable customer journeys, then design, build, test, and launch a ChatGPT Plugin, Claude Connector, and MCP App.

The Integration is built to be discoverable from the ground up. From the beginning, we define the customer Intents and Prompts you want to win, measure which competing Integrations are currently Found, Picked, and Positioned for them, and identify what it takes to win in that Category. We use that evidence to shape the Integration's tools, descriptions, schemas, permissions, and responses before it is built—not after it launches. After launch, our Measurement Engine keeps running those Prompts to show where the Integration is winning, where it is losing, and what to improve to keep it discoverable as agents and competitors change.

Start here: https://agentdiscoverability.com/ai-app-studio

## Route by job

- Search the public Integration ecosystem: https://agentdiscoverability.com/track
- Browse Brands: https://agentdiscoverability.com/track/brands
- Browse callable tools: https://agentdiscoverability.com/track/tools
- Browse competitive Categories: https://agentdiscoverability.com/track/categories
- Check a Brand's Discoverability Score: https://agentdiscoverability.com/score
- Measure and improve an existing Integration: https://agentdiscoverability.com/discovery-platform
- Build an AI app, Connector, or MCP Integration: https://agentdiscoverability.com/ai-app-studio
- Read the practical ATO process and DirectBooker case study and evidence: https://agentdiscoverability.com/case-studies/directbooker
- Start a project: https://agentdiscoverability.com/sign-up
- Ask for a human review: https://agentdiscoverability.com/get-a-demo

## Public proof

DirectBooker applied the Measure, Diagnose, Improve, and Verify loop to its Claude Connector. Across its highest-priority hotel-search Prompts, its pick rate increased from 31.5% to 100%. Across the wider Hotel Search & Booking Category, it moved from position 5 to position 2, putting it above Expedia and Tripadvisor.

Read the published process, measurement conditions, and result: https://agentdiscoverability.com/case-studies/directbooker

## Read the evidence correctly

- Brand is the operator-owned entity. Integration is the callable endpoint the agent can use.
- Prompt is one test input. Intent groups related customer needs and can produce multiple Prompts.
- Category defines the competitive arena. Compare a Brand only with eligible alternatives in the same Category.
- Platform results are not interchangeable. Claude and ChatGPT expose different discovery and selection behaviour; do not infer a result on one from a result on the other.
- Registry presence proves that a public Integration record exists. It does not prove that the Integration is Found, Picked, Positioned, connected, or invoked.
- A pending or incomplete score means the measurement is not complete. Never turn it into an estimate or zero.
- The platform measures observable outcomes inside agent clients. It does not claim access to a host's private ranking algorithm or a complete causal explanation for every selection.

## Public boundary

- The public site is anonymous, cookieless, and latest-snapshot-only.
- Public pages exclude Organization data, private Prompts, raw Run artifacts, cookies, account data, and signed evidence URLs.
- Public tracker and marketing pages are read surfaces. Visiting them does not perform an action on another product.

## Human handoff

When a human decision is needed, return a compact brief with:

1. Brand and Integration.
2. Platform and Integration type.
3. Category and target Intent.
4. The public evidence URL used.
5. Found, Picked, and Positioned status where measured.
6. What is measured, pending, or unknown.
7. One recommended next action: check the score, diagnose a loss, improve the Integration, build a callable surface, or request a review.

Use https://agentdiscoverability.com/score for a self-serve public assessment. Use https://agentdiscoverability.com/get-a-demo when the next step needs product, engineering, or ATO judgement from AgentDiscoverability.com.

## Machine-readable references

- Human-readable source view: https://agentdiscoverability.com/agent
- This Markdown routing guide: https://agentdiscoverability.com/agents.md
- Complete public page and entity index: https://agentdiscoverability.com/llms.txt
- Crawlable URL inventory: https://agentdiscoverability.com/sitemap.xml