Agent visibility leaderboard — 18 sites, May 2026

# Agent visibility leaderboard — 18 sites, May 2026

We scored 18 sites with AgentSite — 7 expected AEO leaders, 7 big-brand SaaS, 4 vibe-coded Lovable builds. The cohort we expected to win didn't. The cohort everyone underestimates beat the household names on median.

By AgentSite · 4 min read · Updated 2026-06-01

We've been saying it for weeks: in the AI-search era, **how an agent reads your site decides whether it cites you**. So this week we stopped saying and started measuring. Eighteen sites. One rubric. Public report for every one.

We split the sample into three cohorts:

-   **Cohort A — expected AEO leaders** (7): Stripe, Anthropic, Vercel, Cloudflare, Mintlify, Cursor, Perplexity. Brands that have publicly invested in agent infra or AEO posture. The ones who _should_ be running the table.
-   **Cohort B — big-brand SaaS** (7): Notion, Linear, Figma, Loom, Slack, Adobe, Datadog. Companies every CMO recognizes. Marketing teams with budget, brand presence that gets cited by ChatGPT every day.
-   **Cohort C — vibe-coded Lovable builds** (4): WanderScout, AgentSwarms, Tymora, GoldMine AI. All AI-vertical SaaS, all featured on _Made with Lovable_. The new wave.

Scoring used the public AgentSite assessment at [agentsite.app/score](https://agentsite.app/score) — up to 8 pages crawled per site, fourteen signals scored across [the five-layer model](/five-layer-aeo), composite 0–100.

## The full leaderboard

| Rank | Site | Cohort | Score | Report |
| --- | --- | --- | --- | --- |
| 1 | [notion.so](http://notion.so) | B | **72** | [report](https://agentsite.app/score/aerp_1kszhpbpm1zm7485yvva8p) |
| 2 | [slack.com](http://slack.com) | B | **71** | [report](https://agentsite.app/score/aerp_1kszhshb50j4f1sppmbbmn) |
| 3 | [cloudflare.com](http://cloudflare.com) | A | **70** | [report](https://agentsite.app/score/aerp_1kszek3vh9m7ge7m4betg6) |
| 4 | [vercel.com](http://vercel.com) | A | **69** | [report](https://agentsite.app/score/aerp_1kszeg9zqft9h8hcyt8vvh) |
| 5 | [stripe.com](http://stripe.com) | A | **66** | [report](https://agentsite.app/score/aerp_1ksze0cmxe9b9rx1b0wb03) |
| 6 | agentswarms.fyi | C | **63** | [report](https://agentsite.app/score/aerp_1kszk9ccn2y705d5cefjv0) |
| 7 | linear.app | B | **62** | [report](https://agentsite.app/score/aerp_1kszhxa4wbd1q55dh35v5e) |
| 8 | [cursor.com](http://cursor.com) | A | **61** | [report](https://agentsite.app/score/aerp_1kszepc3c9c2c4jyadh6g4) |
| 9 | [goldmineai.io](http://goldmineai.io) | C | **56** | [report](https://agentsite.app/score/aerp_1kszje9922d9jcf4dykaqj) |
| 10 | wanderscout.app | C | **55** | [report](https://agentsite.app/score/aerp_1kszjbm009ayqbrjbhe6mh) |
| 11 | [loom.com](http://loom.com) | B | **52** | [report](https://agentsite.app/score/aerp_1kszj1ps0fy6pm2rxs8qm7) |
| 12 | [anthropic.com](http://anthropic.com) | A | **51** | [report](https://agentsite.app/score/aerp_1ksze4avs0ed5rxhzzhq9a) |
| 13 | [mintlify.com](http://mintlify.com) | A | **46** | [report](https://agentsite.app/score/aerp_1kszedg264hfmptp06bevp) |
| 14 | [datadog.com](http://datadog.com) | B | **44** | [report](https://agentsite.app/score/aerp_1kszhgvtaesnh3x6m59xaj) |
| 15 | [tymora.ai](http://tymora.ai) | C | **33** | [report](https://agentsite.app/score/aerp_1kszk74dw012w5rxt3qgch) |
| 15 | [figma.com](http://figma.com) | B | **33** | [report](https://agentsite.app/score/aerp_1kszj71vkfm315v4kv2ww9) |
| 17 | [perplexity.ai](http://perplexity.ai) | A | **21** | [report](https://agentsite.app/score/aerp_1kszhdwakb6a1fqr59zhck) |
| 18 | [adobe.com](http://adobe.com) | B | **19** | [report](https://agentsite.app/score/aerp_1kszj4vk6exs5x5277y77x) |

## The three findings

### 1\. Perplexity scored 21. The answer engine itself.

Lowest in the expected-leaders cohort. The category's posterchild for "AI search should be agent-readable" returns a React SPA shell to anything that isn't a browser. The marketing surface fails the test the product was built to win. This is the [SSR-junk-and-bot-wall problem](/ssr-junk-bot-wall) at its most visible: a brand whose entire thesis is _agents will read the web_ shipping a homepage agents can't read.

### 2\. Notion (72) and Slack (71) lead the entire chart.

Two big-brand SaaS we'd pre-classified as middle-of-pack outranked Cloudflare, Vercel, Stripe, and Anthropic. SEO-heavy SSR is paying off where AEO-leader posture isn't. Notion's content is dense, structured, and reachable without JavaScript. Slack's marketing site renders a real document. Brand recognition isn't what wins; rendered, indexable content is.

### 3\. Vibe-coded beat big-brand on median.

Cohort C median: 56. Cohort B median: 52. AgentSwarms — an AI agent platform built with Lovable — scored 63, higher than Cursor, Anthropic, and Mintlify. The cohort I'd queued up as the "AI invisible to AI" punchline turned out to be the cohort that beat the household names. Lovable's default output for a small app is closer to [agent-readable](/agent-readability) than most enterprise marketing teams are shipping.

## The cross-cohort take

-   Cohort A median (61) leads Cohort B median (52) by **9 points**. AEO posture has measurable value.
-   Cohort C median (56) beats Cohort B median (52) by **4 points**. Brand recognition doesn't predict agent visibility. Willingness to ship a clean rendered surface — and avoid SSR sprawl — does.
-   Within-cohort spread in B is **53 points** (Notion 72, Adobe 19). The variance among big brands is larger than the variance between cohorts.

## Methodology

For each site we ran the AgentSite assessment at [agentsite.app/score](https://agentsite.app/score) on 2026-05-31:

-   **Crawl:** up to 8 pages per site, with the bot-UA / browser-UA fallback that catches edge-blocked roots
-   **Signals (14, across 5 layers):** [llms.txt](/llms-txt) presence and quality, agent-readable HTML access, [structured data and schema fit](/faq-schema), per-page markdown surface, [heading structure](/heading-structure), [direct-answer](/direct-answer) and [definition-density](/definition-density) in the lede, [outbound citations](/outbound-citations) and statistics, and the broader [Layer-1–5 stack](/aeo-problems)
-   **Composite:** weighted 0–100 from the per-layer scores

Every site report is bookmarkable for 30 days at the URL shown in the table above. The methodology is identical to the public score path — anyone can reproduce a row by pasting the URL at [agentsite.app/score](https://agentsite.app/score).

## What's next

This is the first cut of a recurring series. We'll re-run quarterly and post the deltas: who moved up, who moved down, who finally shipped llms.txt. If you're on the list and want a re-run after fixes, send us a note — we'll publish the delta.

If you want the carousel summary, the deck shipped on LinkedIn on 2026-06-01. The methodology is identical; this page is the receipts.

For context on the methodology evolution: the [llms.txt field-report from May 2026](/llms-txt-field-report-2026-05) covers our first probe of the same engine and platform operators — useful baseline reading.

* * *

**Run your own site →** [agentsite.app/score](https://agentsite.app/score) · paste a URL · ~90 seconds · no signup