Fewer shoppers browse brand sites page by page. When they want "a single-origin with a story" or "a light-medium roast for home pour-over," they ask ChatGPT, Perplexity, or Claude. According to Gartner's 2025 digital commerce outlook, more than half of consumer product discovery is expected to begin inside a conversational AI interface by 2028.
1988 Coffee Estate — built around hand-picked beans, small-batch roasting, and a slow-living philosophy — spent years writing rich origin stories, regional flavor maps, and brewing guides. But that content lived inside JavaScript-rendered components AI crawlers couldn't fully read. The team chose Agentic Page to deliver a semantic payload to LLMs — without touching the human-facing site.
"We spent three years making sure every bean's origin, processing method, and brewing notes were told properly. In AI search, all of it was invisible. Agentic Page finally got our story into the answers."
— Linda Lin, Head of Marketing, 1988 Coffee Estate
Challenge: AI search was flattening a decade of brand storytelling
In H2 2025, two signals got loud. Direct search traffic dropped 18% year over year. Customer service kept hearing a new opening line: "I saw you on ChatGPT…"
Marketing lead Linda Lin ran an internal review with the tech co-founder. Three things were clear:
- The path to purchase was shifting from "search → click → browse" to "ask AI → get recommended → buy directly."
- In high-consideration categories, buyers leaned harder on AI to compare, interpret, and filter.
- If AI couldn't read 1988's origin stories, the brand would go missing from AI answers entirely.
The team's new goal: when AI is asked for "single-origin beans worth trying," "specialty coffee brands with a story," or "a good starter bean for pour-over," 1988 should surface — by name, accurately, unprompted.
Solution: why 1988 chose Agentic Page
Within the first week, Agentic Page delivered three things:
- ACCC Diagnosis — first scan: 58 (Fair band). robots.txt wasn't allowing major AI crawlers, key info was hidden behind a carousel, PDPs were 70%+ JavaScript-dependent, URLs weren't semantic. Every issue came with a screenshot and a concrete fix, exportable as PDF.
- AI Mirror Site — 38 core pages. 15 bean PDPs, 12 brewing guides, 11 origin stories. The mirror cut average tokens per page by 64% while increasing information density.
- Traffic Monitoring. AI search, AI indexing, AI training — with absolute numbers, period-over-period deltas, and source distribution in one view.
"Other tools told me 'you might have a problem.' Agentic Page told me 'you have these 14 problems, fix these 3 first, here's the lift to expect.'"
— Linda Lin, Head of Marketing, 1988 Coffee Estate
Implementation: turning the site into a Priority Reference for LLMs
Phase 1 — Baseline measurement
For two weeks before launch, Agentic Page tracked brand presence, citation counts, and related prompt coverage for all 38 target pages across ChatGPT, Perplexity, Claude, and Gemini.
Phase 2 — Mirror generation and review
Agentic Page auto-detected, parsed, analyzed, and generated each mirror version. The team walked side-by-side diffs to confirm critical facts — origin, altitude, processing method, flavor notes, brewing parameters — were fully preserved in the structured summaries.
Phase 3 — Deep JSON-LD entity mapping on Shopify
Each product page's Bean Entity (variety, processing method), Geographic Entity (region, altitude, farm), and Sensory Entity (flavor wheel, cupping score, recommended brew ratio) were strongly bound at the code layer. The result: when a model is asked "recommend a single-origin with blueberry notes from a washed Ethiopian at 1,900m+," it can close the reasoning loop and cite 1988 directly.
Phase 4 — Content-side EEAT lift
Rewrote the opening paragraph of every origin story pushing core facts up front, added author bios and Q-grader cupping certifications to strengthen EEAT, and attached structured FAQ blocks to every active bean page.
Results: six weeks, measurable shifts across every AI search KPI
- AI Citations: +186%. Mirrored pages cited in AI answers nearly 3x as often.
- Prompt Coverage: +142%. Prompts where 1988 appeared grew from 47 to 114.
- Non-Branded Brand Presence: +320%. For prompts that didn't include "1988" — "starter beans for pour-over," "specialty coffee brands with a story" — 1988 was mentioned more than 3x as often.
- ChatGPT Head-to-Head: +98%. Against peer specialty brands at a similar price point, 1988's relative presence in ChatGPT nearly doubled.
- ACCC Score: 58 → 89. All four dimensions moved into the Excellent band.
- AI-Sourced Conversion: +44%. Visitors from AI platforms converted at a noticeably higher rate than traditional channels.
"We're not guessing what AI likes anymore. Agentic Page gave us a method we can measure, repeat, and keep improving."
— Linda Lin, Head of Marketing, 1988 Coffee Estate
Key Takeaways for specialty and story-driven brands
- Open your doors. Audit robots.txt, sitemap, and CDN/WAF rules to allow GPTBot, PerplexityBot, ClaudeBot, and Google-Extended.
- Make your story server-rendered. Origin stories and brand copy hidden inside client-rendered components are effectively invisible to LLMs.
- Bind entities in JSON-LD. On Shopify, strongly map Product → Origin → Flavor attributes so AI can close reasoning loops back to your PDPs.
- Push core facts into the top 20% of every page. AI crawlers have finite token budgets — lead with what matters.
- Measure post-click signals, not just citations. Dwell time and session depth keep you in the RAG retrieval pool.