The situation
Lumē had done everything right by conventional standards.
Their website was clean and fast. Their SEO was in decent shape — ranking on the first page for several category terms. They had a growing organic traffic number, a solid content library, and a founder who understood the importance of digital presence.
But conversions were stubbornly flat. Traffic was coming in. Purchases were not following.
The team's instinct was to look at the usual suspects: the checkout experience, pricing, ad creatives. A couple of A/B tests were run. Nothing moved meaningfully.
What nobody had looked at was what happened before someone arrived at the website — specifically, what AI systems were saying about Lumē when potential customers asked about their category.
What 24LLM found
When the 24LLM audit ran across Lumē's website and cross-referenced how AI systems were interpreting the brand, three things became clear immediately.
The brand was being categorised incorrectly.Lumē positioned itself as a clinical skincare brand — efficacy-first, dermatologist-developed, for people who wanted results over aesthetics. When AI systems summarised the brand, they consistently placed it in the "natural and clean beauty" category. Not wrong exactly, but not right either. The word "clean" appeared prominently on the site in the context of clean ingredients. AI systems latched onto it and assigned a category that attracted a completely different buyer.
People looking for clean beauty arrived. People looking for clinical results — Lumē's actual customer — were not being sent there by AI-powered discovery at all.
The most common buyer doubt was completely unaddressed.Lumē's primary product was a vitamin C serum priced at ₹1,800 — a considered purchase for most buyers. The audit identified that the single most common hesitation in this category, at this price point, was: how long before I see results?
This question appeared nowhere on the website in a clear, direct form. There was a general FAQ section, but the answer was buried under marketing language. An AI system asked this question on behalf of a potential buyer would either skip it, invent a generic answer, or pull a competitor's response from the broader category.
The doubt was there. The answer was not.
Interpretation varied significantly across AI systems.When the same questions were posed to ChatGPT, Perplexity, and Google's AI Overview, the descriptions of Lumē came back meaningfully different. One described it as "a premium Ayurvedic brand." It is not. One said the brand was "popular for oily skin types." Their bestseller is a hydrating serum designed for all skin types, with a particular lean toward dry and combination.
None of these descriptions were accurate. All of them were influencing potential buyers before those buyers ever visited the site.
The findings in plain terms
| What was measured | What was found |
|---|---|
| Category accuracy across AI systems | Incorrectly categorised as "natural/clean beauty" in 4 of 5 AI responses |
| Top buyer doubt coverage | Primary objection (results timeline) not addressed on site |
| AI interpretation consistency | 3 different brand descriptions across 3 AI platforms |
| Narrative control | Core differentiator (clinical, dermatologist-developed) absent from most AI summaries |
| Decision friction | High — buyers receiving incomplete or inaccurate information pre-visit |
What changed
The 24LLM recommendations were specific and prioritised — not a general content audit, but a targeted list of what was actually broken and what fixing it would affect.
Three changes were made in the six weeks following the audit.
Category language was made unambiguous. The homepage, the about section, and the product descriptions were updated to make the clinical positioning explicit and consistent. The word "clean" was kept — their formulations are genuinely clean — but it was contextualised within an efficacy-first narrative so AI systems could not extract it in isolation and draw the wrong category conclusion. The results question was answered, prominently. A simple, direct statement — visible above the fold on the product page — addressed the results timeline question in plain language. Not a promise. A realistic, specific, honest answer. The kind of thing a good salesperson would say if a customer asked in person. The founding story was made consistent across pages. The dermatologist involvement, which was the strongest trust signal in the brand's story, was buried on an about page that most visitors never reached. It was moved to the homepage and the core product pages in a concise, consistent form that AI systems could find and carry forward.What happened after
Lumē did not run new ad campaigns. They did not redesign the website. They did not change their pricing or launch new products.
Eight weeks after implementing the recommendations:
- AI systems consistently categorised Lumée as a clinical skincare brand across all platforms audited
- The brand's description in Perplexity matched its actual positioning for the first time
- Conversion rate on the hero product page increased by 23%
- Average session duration increased — a signal that visitors arriving were the right visitors, not category-mismatched traffic
The traffic numbers did not change dramatically. What changed was the quality of the understanding that preceded each visit.
What this illustrates
Lumée's problem was not SEO. Their rankings were fine. Their technical health was fine.
Their problem was that AI systems — which millions of buyers now use as the first step in any considered purchase — had constructed an inaccurate picture of the brand. That picture was shaping decisions before buyers arrived. And there was no traditional metric that would have surfaced it.
The 24LLM audit found it in a single diagnostic run. The fixes took six weeks. The impact showed up in conversion data two months later.
A note on methodology
All findings in this case study were produced through 24LLM's layered evaluation framework, which analyses a website across six dimensions: knowledge accessibility, answer and belief coverage, AI interpretation stability, narrative and decision control, answer selection readiness, and outcome impact.
AI systems used in the evaluation included ChatGPT, Perplexity, and Google AI Overview. No single model was treated as authoritative. Interpretation variance across systems was a core part of the diagnostic.
24LLM — measuring how well a brand is understood before anyone chooses it.