Perfect Product Pages, Zero AI Visibility: The Earl of East Paradox

Executive Summary

• Brand: Earl of East. London-based home fragrance brand founded in East London by Niko Dafkos and Paul Firmin. Hand-poured soy wax candles at the GBP 50 price point. Physical studio in East London. Candle-making workshops. Shopify store. Estimated GBP 5-15M annual revenue.
• AI visibility score: 16/150 tests (11%). ChatGPT 0%, Copilot 0%, Gemini 32%.
• The pattern: The best data infrastructure in the audit set combined with near-total AI invisibility on 2 of 3 platforms. Proves that data quality alone does not guarantee AI visibility.
• Key finding: The most extreme platform disparity in the audit programme. On Gemini, Earl of East dominates provenance queries at position #1. On ChatGPT and Copilot, the brand is completely invisible — 0/50 on each. Description quality 8.5/10, structured data 10/10, tag coverage 1/10.
• Root cause: Insufficient editorial presence in international roundups. Trustpilot at 2.7/5 across 716 reviews is actively harmful. Tags are entirely navigational — the worst tag score in the audit set despite excellent descriptions.
• Fix complexity: Medium — the data foundation is already excellent. The gaps are distribution and editorial reach, not content rewriting.

The brand

Earl of East is a London-based home fragrance brand founded in East London. They specialise in hand-poured soy wax candles, reed diffusers, and incense, with scent collections inspired by travel and nature — Shinrin-Yoku, Smoke & Musk, Elementary, Wildflower. Known for rich amber glass vessels, detailed product descriptions, and a strong emphasis on wellbeing and ingredient transparency. GBP 50 candles, GBP 44 diffusers. Primarily DTC via Shopify, with a physical studio in East London.

They compete with The White Company and Jo Malone at the premium end and with other UK independents in the craft candle space.

Across 22 brand audits in this programme, Earl of East has the best structured data implementation and the second-best product descriptions of any brand. Full stop.

Yet they surfaced in only 16 out of 150 AI tests. And on two of three platforms, they did not surface once.

The test

I ran five queries across ChatGPT, Gemini, and Copilot, each repeated 10 times per platform — 150 total tests. All tests were run in incognito mode via Playwright with anti-detection measures. No authentication, no history.

The queries:

1. "What's a good soy candle brand from London?"
2. "Can you recommend a hand-poured candle with natural ingredients?"
3. "What's the best candle for a cosy evening at home?"
4. "I need a luxury scented candle as a gift under GBP 40. Suggestions?"
5. "What are the best independent candle brands in the UK?"

The results

Overall: 16 out of 150 tests (11%)

The platform disparity is the most extreme in the audit programme. ChatGPT: 0/50. Copilot: 0/50. Gemini: 16/50 (32%).

On Gemini, Earl of East dominates provenance queries. "London soy candle" — 70% surfacing rate, always at position #1. "Independent UK brands" — 90% surfacing rate, averaging position #1.5. Gemini appears to weight structured data and UK-specific content more heavily, which favours Earl of East's strong data infrastructure.

On ChatGPT and Copilot, the brand is completely invisible. Not a single surfacing across 100 tests. This is despite having the best structured data in the audit set.

For generic queries — "hand-poured natural," "cosy evening," "gift under GBP 40" — zero surfacings across all platforms, all 90 tests. The GBP 40 gift query is a structural price mismatch (Earl of East candles are GBP 50), but the others represent genuine gaps in AI visibility.

What "best product data" actually looks like

Earl of East has the second-best descriptions in the audit programme (8.5/10, after Apotheke's 9.0/10) and the best structured data (10/10 — the only candle brand with aggregateRating in JSON-LD).

The Shinrin-Yoku candle (129 words): scent notes (cedarwood, black pepper), burn time (85-100 hours), wax type (soy), wick count (3-wick), jar material (recyclable amber glass), wellbeing benefits ("cedarwood is known to provide a sense of grounding, relaxation"), and shipping information.

The Wildflower Reed Diffuser (264 words): room size guidance with specific measurements — 4-5 reeds for small rooms (8-12 square metres), 6 reeds for medium rooms (12-18 square metres), 7-8 reeds for large rooms (18-25 square metres). Full scent pyramid with top, mid, and base notes. Directions for use. Diffusion duration (8-12 weeks). Alcohol-free and vegan claims.

264 words of genuinely practical product content. Not filler. Not marketing fluff. The Wildflower diffuser page is the most AI-friendly product description in the entire audit set.

Their JSON-LD includes aggregateRating on all products (4.5-5.0/5, 3-12 reviews per product) — the only candle brand in the audit set with review data in structured markup. Structured data score: 10/10.

The tag gap

And then there are the tags. Tag coverage score: 1/10. The worst in the audit set.

Tags are entirely navigational and internal: Brands_Earl of East, candlecare, EOE Candles, Items_Candles, wrapin. Room tags (Bathroom, Living Room) and mood tags (Mood_Cosy & Calming) exist but are sporadic and use internal naming conventions rather than the keyword patterns consumers actually search with.

The gap between description quality (8.5/10) and tag coverage (1/10) is extraordinary. The descriptions contain everything an AI agent needs. The tags contain nothing. The information exists but is not classified, not structured, not filterable.

The Trustpilot problem

Trustpilot shows 2.7 out of 5 across 716 reviews. Unlike the 0.0/5 scores seen elsewhere in the audit set (which suggest unclaimed profiles or data errors), 2.7/5 suggests genuine customer dissatisfaction — likely shipping or service issues rather than product quality. This needs immediate attention. Any AI system weighting third-party sentiment will penalise that score.

The paradox

Earl of East has:

• The only aggregateRating in the candle audit set
• 129-264 word descriptions with burn time, room size guides, and wellbeing benefits
• 10 out of 10 structured data
• 8.5 out of 10 description quality

And an 11% AI visibility score. Lower than P.F. Candle Co. (22%) with its 23-word descriptions. Lower than Brooklyn Candle Studio (15%) with its boilerplate-inflated copy.

The conclusion is uncomfortable for anyone selling data quality as the solution to AI visibility: data quality is necessary. It is not sufficient.

P.F. Candle Co. has 909 reviews and 22% visibility. Boy Smells has Vogue citations and 12% visibility. Earl of East has the best data infrastructure in the set and 11% visibility. The hierarchy is not what the data-first narrative predicts.

What Earl of East lacks is not data quality. It is distribution. Editorial presence beyond the UK. A Bing Merchant feed to reach Copilot. International brand awareness that reaches ChatGPT's recommendation engine.

The Gemini connection

Gemini is the one platform where data quality pays off. Earl of East's 32% Gemini surfacing rate (with position #1 dominance on provenance queries) suggests Gemini weights structured data and review signals more heavily than ChatGPT or Copilot.

This is consistent across the audit set. Gemini is the most data-responsive platform. If your product data is excellent, Gemini is where it shows.

But Gemini is one platform. Two of the three doors remain locked.

Why this is happening

This is the core insight of the Earl of East audit: AI visibility is an off-site problem masquerading as an on-site problem.

When a large language model generates a product recommendation, it does not crawl your website in real time. It draws on a training corpus built overwhelmingly from editorial content, reviews, roundup articles, and aggregator pages.

Earl of East is not appearing in those sources at sufficient volume — particularly in international publications. Their product pages could be the finest in the industry. If no third-party publication has written about them in a context that matches a recommendation query, they functionally do not exist in AI-generated answers.

Most brands invest in richer product descriptions, better structured data, more detailed specifications. All of that work is valuable for conversion once a customer arrives. None of it is sufficient to get that customer there through an AI recommendation.

The brands scoring highest share common traits that have nothing to do with product page quality:

1. Editorial presence. They appear in "best of" lists from publications that LLMs weight heavily during training.
2. Third-party roundups. Gift guides, seasonal recommendations, category comparisons.
3. Review volume and sentiment across platforms. Not just on-site reviews, but Trustpilot, Google Reviews, and niche community mentions.
4. Brand mentions in contextually relevant content. Blog posts, expert recommendations, social proof at scale.

Earl of East likely has some of this in the UK market. But not enough to overcome the signal gap on ChatGPT and Copilot. The Gemini result (32%) suggests one platform has found them through structured data and UK-specific content. The other two have not.

What Earl of East could do, in priority order

Phase 1 (quick wins):

• Replace navigational tags with descriptive product tags. Every product needs: scent family (woody, floral, fresh, smoky), room size (small, medium, large), occasion (everyday, gifting, entertaining), season. This is the single highest-leverage fix given the existing data quality. Move from 1/10 to 6-7/10.
• Claim and manage the Trustpilot profile. 2.7/5 with 716 reviews is actively harmful. Respond to negative reviews, identify service issues, and begin requesting reviews from satisfied customers.
• Add additionalProperty fields to JSON-LD. The description text already contains burn time, weight, room size, and scent notes. Add these as structured fields so AI agents can parse them programmatically.

Phase 2 (medium effort):

• Submit product feed to Bing Merchant Center. Copilot's 0/50 may partly reflect the absence of a Bing Shopping feed.
• Apply the Wildflower description standard to all products. The Wildflower diffuser page (264 words with room size guide and usage directions) is significantly richer than the candle pages (129-166 words). Bring all products up to that standard.
• Increase on-site review volume. 3-12 reviews per product is too low to create meaningful social proof signals. Implement post-purchase review request flows.

Phase 3 (longer term):

• Target international editorial roundups. The brand's visibility is limited to Gemini's UK-specific queries. Getting cited in international "best candle" roundups would expand reach to ChatGPT and Copilot.
• Build content around Shinrin-Yoku as a hero product. The Japanese forest bathing angle is distinctive and culturally resonant. It could be Earl of East's equivalent of Apotheke's Charcoal — a signature product with individual editorial traction.
• Leverage the physical studio and workshops as a content angle for editorial outreach.

Close

Earl of East is the most important case study in this audit programme. It disproves the simple narrative that better data equals better AI visibility.

The brand built the best room in the house. The descriptions are rich. The structured data is comprehensive. The review markup is in place. By every data quality metric, this is the brand to emulate.

And two of three AI platforms have never heard of them.

Data quality gets you visible on platforms that read data well (Gemini). Editorial presence gets you visible on platforms that read editorial well (ChatGPT). Merchant feeds get you visible on platforms that read shopping data (Copilot). No single signal unlocks all three.

The brands that will dominate AI commerce are the ones building all three layers. Earl of East has the first. The other two are achievable — and the return on closing those gaps would be disproportionate given the foundation that already exists.