About

Moss Agents measures whether AI assistants name a brand when a buyer asks about its market — and publishes its own measurement first.

Last updated 18 September 2026

Who runs this

Moss Agents is built and operated by Shiliang — one independent developer, not an agency, not a funded startup with a team behind a landing page. There is no office, no sales department, and no investor to report to.

That is worth knowing before you buy, and it cuts both ways. You get the person who writes the code answering your email, and changes land in days rather than quarters. You also get one person's capacity, which is why the roadmap is short and carries no dates. Payments are taken by Paddle as merchant of record, so your counterparty for billing is an established payment company, not an individual.

Why it exists

Buyers now ask an assistant which tool to use and act on the answer without opening a search results page. That shift is easy to describe and hard to see: there is no ranking to refresh, and the answer is different every time you ask.

Measuring it properly is mostly a discipline problem rather than a modelling one: ask real buyer questions, ask them more than once, store every answer, and refuse to compare two scans that were not taken the same way. That last rule is where most of the work went, and it is the reason the report shows an interval around the score instead of a single confident number.

The claim you can check

The AI visibility index is measured with the same pipeline that scans a paying customer's site. Our own site is in one of the categories we monitor, and its entry is not a flattering one — you can look at exactly where we stand, in public, next to our competitors.

We put it there for a reason: a vendor selling visibility measurement while hiding its own measurement is asking you to believe the thing they refuse to demonstrate. If our method flatters us or buries us, you can see it happening instead of taking our word for it.

The full method — formula, weights, sampling, uncertainty, and the parts we know are weak — is published for the same reason. So is the fact that on an engine which does not browse the web, one component of the score is unreachable for everyone.

What we are not

  • Not an agency. We do not sell content writing, link building or "AI SEO" retainers. The product tells you what is missing; the work stays yours, and if you want it done for you, an agency is the right purchase and we will not pretend otherwise.
  • Not a data broker. Your projects are not sold, and they are not used to train anything. What we send to model providers, and why, is set out on the privacy page.
  • Not a source of testimonials. There are no customer quotes on this site, because there are not yet enough customers for a quote to mean anything, and invented ones are the cheapest lie a company like this can tell. When there are real ones, they will name the customer and be checkable.

How we handle being wrong

Every page that makes a claim about how the product works is written from the code, and the code ships through a test suite that must pass before deployment. That is not a guarantee — this project has shipped three sentences that were true when written and false a week later, and each was caught by checking the live page rather than by the tests.

So if you find a claim on this site that the software does not honour, that is the most useful thing you can send us. Refunds do not require you to justify them, and a wrong claim is a bug worth reporting whether or not you are a customer.

Get in touch

Questions about this page, or about your data? Write to mossgame@163.com. We answer from a monitored inbox, and billing questions are answered by Shiliang.

Data requests, refund requests, security reports and "this number looks wrong" all reach the same person.