We audited our own store for AI search and published the score.
63 out of 100. The store had a thorough llms.txt file and a homepage that introduced itself as "Agril". Both things were true at the same time, one file apart.
- Business
- AgrilHoTech, our own e-commerce store
- Sector
- Premium houseplants and tropicals, shipping island-wide in Sri Lanka
- Stack
- WooCommerce on WordPress, LiteSpeed hosting
- What we did
- Built the store, run the marketing, and ran a scored AI search visibility audit against it
- Live at
- agrilhotech.com
- Baseline
- 63/100 (Discovery 82, Data quality 48, Actionability 60)
Why our own store is on this page
Every agency has a client list. Very few have a business of their own that has to survive the advice they sell. AgrilHoTech is ours. It carries real stock, real shipping, real customers and real consequences, which makes it the only place we can test something properly before charging anyone for it.
Publishing a starting score of 63 out of 100 on our own property is deliberate. A baseline you can go and check is worth more than a result you cannot.
The problem worth solving
Search is no longer only a page of ten blue links. A meaningful share of buying questions now get answered by an assistant that reads the web and replies in a sentence. "Where can I buy an anthurium in Sri Lanka" is exactly that kind of question. If a store is not legible to those systems, it does not appear in the answer, and there is no ranking report to tell you it happened.
So we built a scored rubric: how discoverable the site is, how good its structured data is, and how actionable the information is once found. Then we ran it against our own store.
What the audit found
The good half was genuinely good. robots.txt explicitly welcomes GPTBot, PerplexityBot and ClaudeBot while blocking the scrapers that take without returning anything. The llms.txt file was thorough. Product pages carried real structured pricing and a substantial FAQ block. Discovery scored 82.
The bad half was embarrassing and cheap to fix. The homepage title read "Home - Agril". The shop page description was the plugin's own unedited placeholder text, shipped exactly as installed. The organisation data was missing its logo, its social profiles, its contact point and its address. Data quality scored 48.
Where the two halves connected
Look at what fixing this requires. robots.txt is a server-level file. Structured data lives in the theme and the SEO plugin. Page titles come out of a template. None of those are things a marketing agency can edit, and all of them were being changed for a purely marketing reason: being present when a buyer asks an assistant where to shop.
A development team would not touch robots.txt on behalf of answer engines, because nobody would have asked them to. A marketing team would have written the recommendation and waited. The work happened because one team held both the hosting credentials and the reason.
The result
Baseline: 63/100, from a live audit run on 12 July 2026 against a fixed three-part rubric, with the evidence file saved. Discovery 82, data quality 48, actionability 60.
The rubric is fixed on purpose so the re-audit is a comparison rather than a fresh opinion. This is the same method we run for clients, which is why we ran it on ourselves first.
What we would do differently
Read the rendered page titles before writing the llms.txt. The order was backwards. The advanced work was done while a default placeholder sat on the most important page on the site. It is a good reminder that sophistication at one layer hides nothing about the layer beneath it, and that the cheapest wins are usually the ones nobody thought to check.