We Ran an AI Readiness Audit on a D2C Brand. It Scored 46 Out of 100.
Ask ChatGPT, Claude, Gemini, or Perplexity a question about a D2C brand today - "Is this product good for sensitive skin," "What's the return policy," "Is this brand legit" - and you'll get an answer. The question is whether that answer is actually right, and whether it's built from your brand's own information or from whatever the model could scrape together.
We started running a structured check on this, and the results have been more useful than we expected. We call it an AI Readiness Audit. In July, we ran one on Pokonut, a real D2C skincare brand, and it scored 46 out of 100. The result was labeled "at risk."
That number is not a vanity score. It's built from two separate checks that most brands have never had run on them.
The first check: what do the models actually say
We ask four AI assistants a set of real shopper questions about the brand - the kind a customer would actually type before buying. Then we grade each answer against the brand's own site and product pages. Is the answer accurate? Is it current? Does it cite the brand's actual claims, or does it invent something plausible-sounding? Models are fluent even when they're wrong, and a confidently wrong answer about ingredients, pricing, or returns is worse for a brand than no answer at all.
The second check: can the models even read your site properly
This is the part most marketing teams have never looked at, because it's not a marketing problem in the traditional sense - it's closer to infrastructure. We check:
robots.txt policy toward AI crawlers (many sites unknowingly block the exact bots that would answer questions about them); sitemap.xml presence and accuracy; llms.txt, a newer convention some sites use to describe themselves directly to language models; whether the server sends real HTML to a non-JavaScript agent, or just an empty shell that only renders after JavaScript runs; whether a products.json or equivalent feed is open and current; JSON-LD structured data on product pages, the schema markup that tells a crawler "this is a Product, this is its price, this is its rating" in a machine-readable way, not just visually on the page; emerging agentic commerce endpoints (MCP, UCP) that let AI agents transact directly; and general feed hygiene - stale prices, broken variants, missing availability.
Most D2C sites fail several of these without anyone on the team knowing it. It's not usually negligence. It's that these checks sit in a gap between engineering and marketing, and nobody owns the gap.
Why this actually matters commercially
The instinct is to file this under "interesting, not urgent." We'd push back on that. Two things are happening at the same time: shopping behavior is shifting toward asking an assistant before buying, especially for considered purchases, and the assistants doing the answering can only be as accurate as what they can actually read from the brand. A site that's invisible or unreadable to these tools isn't just missing an SEO checkbox - it's losing the ability to be represented accurately at the exact moment someone is deciding whether to trust it.
We found this out the direct way. We run niti.ai itself, and a similar audit on our own site turned up a real gap: it was a client-side-only application with nothing meaningful in the raw HTML - no title, no description, no structured data - until JavaScript executed. Any crawler or agent that doesn't run JavaScript saw essentially a blank page. We fixed it with a static fallback that serves real content to bots while still giving real visitors the fully interactive version. It's the kind of fix that's invisible to a customer and completely decisive to a model deciding what to say about you.
What a score like 46/100 actually tells a founder
The Pokonut score wasn't a single failure, it was a stack of small ones: a couple of inaccurate AI answers, gaps in structured data, and a technical crawl issue that meant part of the catalog wasn't reliably visible to non-JS agents. None of these are hard to fix individually. None of them get fixed if nobody measures them.
If you want to know where your brand actually stands, this is not a project. It's a few hours of structured checking against your live site, and it comes back as a scored, specific list - not a vague "improve your SEO" recommendation. We're happy to run this for any D2C brand that wants a real number instead of a guess. No pitch attached to the offer - it's useful on its own, whether or not you ever talk to us again after seeing it.