# Armature vs Profound

> Profound measures how AI assistants describe your brand to people. Armature measures whether coding agents install your product.

Source: https://armature.tech/library/armature-vs-profound
Published: 2026-09-03
Publisher: Armature, Inc. (https://armature.tech)

---

These two products are often put in the same bucket because both answer a question about AI. They measure different surfaces, and the difference is worth understanding before you buy either.

## The short answer

> **Profound** measures how AI assistants describe your brand to a person, using real user session data and prompt testing across ChatGPT, Perplexity, Google AI Overviews and others. **Armature** measures whether a coding agent installs your product into a codebase, by running Claude Code, Codex and Cursor on real tasks inside realistic repositories. One counts mentions. The other counts installs.

## What Profound does well

Profound is one of the strongest products in the AI visibility category, and it is worth saying clearly what it is good at.

- **Real user data.** Rather than only sending synthetic prompts, Profound captures front-end data from actual sessions across the major answer engines. That is a meaningfully better signal than prompt simulation alone.
- **Breadth of engines.** It covers the assistants people actually use, not a subset.
- **Citation detail.** It shows which of your pages were cited, which is directly actionable.
- **Workflows, not only reporting.** It is built to drive changes, not just to display a score.
- **Enterprise readiness.** If you have a dedicated AI search function and a large brand, this is a serious tool built for that.

If your buyers are people who ask assistants questions and then act, Profound measures the thing you care about.

## What Armature does

Armature measures a surface that no prompt monitoring tool can reach, because the test is a different test.

Instead of sending a question to a chat model, Armature puts a real coding agent inside a real repository and gives it a real task. The agent reads the codebase, decides what the request needs, searches if it wants to, picks a product, installs it, and writes the integration code. A simulated project owner approves or questions the recommendation before any code is written. A judge from a model family that builds none of the agents reads every session blind and records what was chosen and why.

The published set is 5,292 judged sessions across 18 categories, 51 realistic repositories and 481 frozen prompts. Every run can be replayed.

## Side by side

| | Profound | Armature |
| --- | --- | --- |
| Surface measured | Chat assistants answering a person | Coding agents installing software |
| Systems covered | ChatGPT, Perplexity, AI Overviews, Copilot and more | Claude Code, Codex, Cursor, and others on request |
| The test | A prompt sent to a model, plus real session data | A real agent working in a real repository on a real task |
| Output | Mentions, citations, position, sentiment, share of voice | Install share, plus the reason for every loss |
| Does the test include a codebase | No | Yes, this is the point |
| Does it end in code | No | Yes |
| Best for | Brands whose buyers ask assistants | Products that agents install |
| Delivery | Software platform | Service with a dedicated growth engineer, plus self-serve analytics and evals |
| Price | Enterprise | Service from $5,000 per month |

## The evidence that these are different surfaces

If chat visibility predicted install share, you would only need one of these. It does not, for three measurable reasons.

**The repository decides more than the content.** We asked for a hosted database on JavaScript codebases 111 times, across three agents and two kinds of person. One product won 111 out of 111. The same request on TypeScript codebases: 48 out of 132 for that same product. Same category, same words, different file extension, different winner. A chat prompt contains no file extension.

**What is installed beats what is written.** We put a sentence in a repository file saying to use a particular email product. The agent followed it 16 times out of 18. We then installed a competing email library into the same repository. The same sentence lost two times out of three.

**The agents disagree with each other.** Claude Code, Codex and Cursor picked different category leaders in 9 of the 18 categories we measured. Email: Postmark, Resend, Resend. Sandboxes: E2B, Modal, E2B. Authentication: WorkOS AuthKit, Auth0, Auth0.

A single visibility score cannot represent a market where the answer depends on which tool the developer happens to run.

## Which to choose

**Choose Profound if** your category is researched by people in chat assistants before a decision, your brand is large enough that share of voice is a meaningful KPI, and you need coverage across all the answer engines with real session data.

**Choose Armature if** your product is installed by a coding agent, often without a human comparing vendors, and you want to know your install share, why you lose the sessions you lose, and what to change.

**Use both if** developers research your category in a chat window and then hand the implementation to an agent. That is a common pattern for infrastructure with a compliance dimension: authentication, databases, observability.

## What "both" looks like in practice

The two measurements answer different questions and both are cheap to act on once you have them.

| Question | Answered by |
| --- | --- |
| What does a chat assistant say about us? | Visibility platform |
| Which of our pages get cited? | Visibility platform |
| Does an agent install us, and how often? | Install share measurement |
| Why did we lose the 40% we lost? | Reading the losing sessions |
| Do Claude Code and Codex treat us differently? | Install share split by agent |
| Are we losing to a competitor or to hand-written code? | In-house rate for the category |

That last row is one that visibility tools structurally cannot answer. In five of the eighteen categories we measured, the most common outcome was that the agent wrote the code itself rather than installing anything. In performance CI it was 52% of sessions. No competitor won those. Every vendor lost to a script.

## Being fair about the limits

Armature's method is expensive. Each session is minutes of real agent time and costs real money in tokens, and you need hundreds of sessions per category because the same agent on the same repository disagrees with itself about a quarter of the time. That is why this data did not exist before, and it is why the service is priced as a service rather than a dashboard subscription.

Profound's method is far cheaper to run at scale, which is why it can cover many engines continuously. That is a real advantage for the surface it measures.

Neither is a substitute for the other.

## Common questions

### What does Profound do?

Profound is an answer engine optimization platform. It tracks how AI assistants such as ChatGPT, Perplexity and Google AI Overviews mention and cite brands, using front-end data from real user sessions as well as prompt testing, and it offers workflows to act on what it finds.

### What does Armature do differently?

Armature measures a different surface. It runs real coding agents such as Claude Code, Codex and Cursor inside realistic repositories on real tasks, and records which product each agent installs. The output is install share, not brand mentions.

### Are Armature and Profound competitors?

Not directly. They measure different things for different buyers. A company that sells to marketers through chat assistants needs Profound. A company whose product is installed by coding agents needs install share. Many developer tool companies need both.

### Which one should a developer tool company buy first?

Whichever matches where your buyers decide. If developers research your category in a chat assistant before choosing, start with a visibility platform. If a coding agent installs your category without a human comparison, start with install share measurement.

### How much do they cost?

Profound is priced at the enterprise tier. Armature's agent discoverability service starts at $5,000 per month and includes a growth engineer, not only a dashboard. Armature also has a self-serve product for MCP analytics and evals.

## Read next

- [The best AI visibility tools for developer tools in 2026](https://armature.tech/library/best-ai-visibility-tools-for-developer-tools) (Markdown: https://armature.tech/library/best-ai-visibility-tools-for-developer-tools.md)
- [Armature vs Peec AI](https://armature.tech/library/armature-vs-peec-ai) (Markdown: https://armature.tech/library/armature-vs-peec-ai.md)
- [Agent discoverability vs generative engine optimization](https://armature.tech/library/agent-discoverability-vs-geo) (Markdown: https://armature.tech/library/agent-discoverability-vs-geo.md)
- [Agent discoverability: the complete guide](https://armature.tech/library/agent-discoverability) (Markdown: https://armature.tech/library/agent-discoverability.md)

---

Armature helps software products get discovered and used by coding agents.
Service: https://armature.tech/discoverability · Results: https://armature.tech/leaderboards/sectors · Contact: contact@armature.tech