# Benchmarking your agent readiness against competitors

> How to benchmark agent readiness against competitors fairly: AgentScore on each site, categories not totals, store study context and hand checks.

By Ghost Agent Lab · Measurement · https://ghostagentlab.com/articles/benchmark-competitors-agent-readiness/

## Key takeaways

- A benchmark against your direct competitors shows how urgent your agent readiness gaps are, in terms leadership teams understand.
- Compare category scores and individual checks, not just totals, and only compare checks that ran on both sites.
- In our study of 255 online stores, the median store scored 80 and earned 72% of the available navigability points, the weakest category.
- Keep competitor checks to a free scan and ordinary shopping by hand: no load testing, automated checkout runs, fake accounts or test orders.
- This week, scan your site and three competitors with AgentScore on the same day and list every check you fail that a competitor passes.

When an AI assistant compares stores for a shopper, it compares you with your competitors. Here's how to benchmark your agent readiness against theirs fairly and safely: running AgentScore on each site, comparing categories rather than totals, using our store study as context, checking journeys by hand, and turning the results into a plan.

## Why benchmark

Your own AgentScore tells you what to fix. A benchmark tells you how urgent it is. If a shopper asks an assistant for "a waterproof jacket under $200 that can arrive by Friday", the assistant can only weigh the stores whose prices, stock and delivery it can read, and only help buy where it can reach the checkout. Being behind your direct competitors on those things matters more than being behind an abstract ideal.

A benchmark is also the easiest way to explain agent readiness to people who don't follow it closely. "We score lower than two of our three main competitors on navigability" is a sentence a leadership team understands. Our guide to [explaining agent readiness to your board](https://ghostagentlab.com/blog/explain-agent-readiness-to-your-board/) has more on framing it.

1. **Choose:** A small set of comparable competitors
2. **Scan:** Each site with AgentScore, on the same day
3. **Compare:** Category by category, then check by check
4. **Check by hand:** A few journeys, as an ordinary shopper
5. ✓ **Act:** Fix the gaps that matter most, then repeat

**A benchmark in five steps.** The scan gives you numbers; the comparison and the hand checks tell you what they mean.

## Choose who to compare

Pick a small set, three to five sites, that a shopper might realistically weigh against you:

- **Direct competitors** selling similar products to similar customers.
- **One store you admire**, even if it's bigger, to see what good looks like in your category.
- **Like for like.** Compare stores with stores and software sites with software sites. AgentScore runs store-only checks, such as guest checkout and product data, only where it finds products or a cart, so a store and a non-store aren't scored on the same checks.

Use the address customers actually visit, such as the www version if that's where the home page lives, and the same country site for every competitor where you can. Pop-ups and content that vary by country can change what a scan sees.

## Run AgentScore on each site

[AgentScore](https://ghostagentlab.com/agentscore/) is free and needs no account: enter each address and your email, and you get a score out of 100, a score for each category, and a pass, warning, fail or "not tested" for every check. The report has its own link, so save it with the date.

A few things to know so you use it fairly:

- **The free page scans each site at most once a day.** It reuses a domain's result for 24 hours, so repeated requests don't hit that site again. If you scan a competitor twice in a day, you'll see the same result. That's deliberate, and it's good manners: there's no reason to scan anyone more often than that.
- **The free page also limits how many new scans one visitor can start in an hour.** Benchmarking a handful of sites is well within normal use. Scanning a whole market in bulk isn't what the free page is for.
- **Sites can opt out, and you should respect it.** A site can ask AgentScore not to scan it, with a rule in its [robots.txt](https://www.rfc-editor.org/rfc/rfc9309.html) or by asking us directly. If a competitor has opted out, the scan stops after reading robots.txt and tells you so. Leave that competitor out of the scanned comparison. The [AgentScore bot page](https://ghostagentlab.com/agentscore/bot/) explains how the scanner identifies itself and how opting out works.
- **Scan everyone close together.** We change how AgentScore scores from time to time, and scores from different scoring versions aren't comparable. Scan every site on the same day, and re-scan all of them when you repeat the benchmark.

The scan is light by design. It reads public pages, visits at most about 20 pages, and never fills in or submits a form or places an order. The only things it presses are a product's add-to-cart button, once, to see the checkout, and a pop-up's close button if one covers it. [How AgentScore works](https://ghostagentlab.com/blog/how-agentscore-works/) explains every check and how the points add up.

> **Scan, don't probe.** AgentScore is a fair way to see what AI agents see on a competitor's public pages. Don't go beyond it: no load testing, no automated agents running through a competitor's checkout, no fake accounts or test orders. Besides being unfair to them, it may break their terms of use, and it can trip their bot protection in ways that distort what you learn.

## Compare categories, not just totals

Two sites can both score 80 for very different reasons. One might let every agent in but have unnamed buttons everywhere; the other might be easy to use but block AI assistants at checkout. The total hides the difference, and the difference is what tells you what to do.

Put the results in a simple table, with the category scores each report shows:

| Site | AgentScore | Access (of 25) | Readability (of 25) | Navigability (of 20) |
| --- | --- | --- | --- | --- |
| Your store |  |  |  |  |
| Competitor A |  |  |  |  |
| Competitor B |  |  |  |  |
| Competitor C |  |  |  |  |

Then go one level down. For each check, mark where you fail and a competitor passes. Those are your clearest gaps. Two cautions:

- **"Not tested" isn't a pass.** Checks that don't apply to a site are left out of its score, never counted against it, so a competitor whose checkout the scanner couldn't reach may look better on paper than it is. (A check the site itself stopped from running, shown as "couldn't test", does cost points.) Compare only checks that ran on both sites.
- **Task completion isn't in the free score.** The free scan doesn't send an AI agent through a real task, so the scores cover Access, Readability and Navigability. That's why the hand checks below matter.

Our guide to [reading your AgentScore report](https://ghostagentlab.com/blog/reading-your-agentscore-report/) explains what each result means and what to fix first.

## Use the store study as context

A wider reference point helps when your competitor set is small. In [our study of 255 online stores](https://ghostagentlab.com/blog/agentscore-store-study-2026/), scanned on October 20, 2026 in scoring version 2026.10-v8: "The median store scored 80 out of 100. 51% scored 80 or more, and none scored below 50."

**Median store's share of available points, by category**

- Access: **92%**
- Readability: **94%**
- Navigability: **75%**

**Where the median store stands.** From our study of 255 stores: getting in and reading pages is mostly solved, and navigability is where agents struggle.

By category, the study found that "the median store earned 92% of the available points for access, 89% for readability and 72% for navigability." To compare your own category scores, divide by the category's points: 20 out of 25 for access is 80%, below the median store's 92%.

Keep the study in its place. It's a sample drawn from the online shops listed in Wikidata, which leans toward established retailers, scanned from a single location on one day, and its figures only apply to scans made with the same scoring version. The free AgentScore report shows how a score compares with the study when the versions match.

## Check journeys by hand

A scan can't tell you whether an agent can actually buy. A short hand check can give you a feel for it, as long as you do it the way any shopper would:

1. **Pick two or three tasks** that matter in your market, such as "find a size 10 trail shoe in stock and see the delivery cost" or "find the price of the mid-tier plan".
2. **Do each one by hand** on your site and each competitor's, as an ordinary visitor. Stop before you pay or submit anything. Note the steps, and anywhere a button had no name, a pop-up got in the way or a price only appeared late.
3. **Try the keyboard.** Walk the same journey with the Tab key. Where you get stuck, an agent probably does too. Chrome's [DevTools accessibility reference](https://developer.chrome.com/docs/devtools/accessibility/reference) shows how to see the names and roles agents read.
4. **Ask assistants the same questions.** Ask a few consumer AI assistants what each store sells, what it costs and how fast it delivers. Note who gets named and whether the answers are right. In the Ghost Agent Lab app, AI visibility (in beta) tracks how Claude answers questions like these, for you and up to three competitors: see [tracking how AI assistants describe your brand](https://ghostagentlab.com/articles/brand-answers-ai-assistants/).

For your own site, go further than a hand check: [test your key journeys with AI agents](https://ghostagentlab.com/articles/test-journeys-with-ai-agents/), using Ghost Agents on a schedule. Keep automated testing to sites you run.

## What to do with the results

- ✓ **List your gaps:** Every check you fail and a competitor passes.
- ✓ **Fix Access first:** If agents can't get in, nothing else matters.
- ✓ **Then the biggest category gap:** Usually where a competitor is clearly ahead.
- ✓ **Repeat on a schedule:** Same competitors, same scoring version, every quarter.
- ! **Don't chase the total:** A higher number matters less than closing the gaps shoppers will feel.

**Turning a benchmark into a plan.** The gaps tell you what to fix; repeating the benchmark tells you whether it worked.

Share the benchmark with whoever owns each fix: most Access problems belong to whoever runs your CDN or bot protection, and most Navigability problems to developers. Then report progress alongside the measures in [agent readiness KPIs](https://ghostagentlab.com/articles/agent-readiness-kpis/). A benchmark that's repeated with the same competitors and the same scoring version is one of the clearest ways to show the work is paying off.

## Sources and further reading

- [RFC 9309: Robots Exclusion Protocol](https://www.rfc-editor.org/rfc/rfc9309.html) (IETF)
- [Accessibility features reference](https://developer.chrome.com/docs/devtools/accessibility/reference) (Chrome for Developers)