Most merchants treat GEO as a one-week “fix the schema” project. The stores that actually move citation share treat it as a 90-day program with three phases, weekly rituals, and measurable checkpoints. Here's the exact blueprint we run with Surfient customers — the deliverables, owners, and numbers to hit at Day 30, Day 60, and Day 90.
Why 90 days and not 30
AI crawlers don't refresh like Googlebot. GPTBot and PerplexityBot hit a typical Shopify store every 9–14 days for PDPs and every 4–6 weeks for collection pages. Even if every schema fix shipped perfectly on Day 1, you'd need two full crawl cycles before the index reflected it. Fewer than 30 days isn't impatience — it's mathematically too short to see signal.
Ninety days gives you three crawl cycles, two rounds of assistant retraining cadence, and enough weekly measurement to separate noise from trend. That's the reason this blueprint has three 30-day phases — Audit, Fix, Amplify. Collapsing any of them costs more than it saves.
Phase 1 · Days 1–30 · Audit
Audit is the phase merchants skip most often and regret most often. The goal isn't to fix anything yet — it's to know exactly what the AI index currently thinks of you, what's broken, and which competitors are eating your citations. Without this baseline every later “it's working” or “it's not” is a guess.
The five deliverables
- Crawler-log baseline — 30 days of server logs, counting hits from GPTBot, ChatGPT-User, PerplexityBot, ClaudeBot, Amazonbot, Applebot-Extended. Record fetched URL, status, user-agent, timestamp. This becomes the truth source, not GA4.
- Citation-share panel — assemble 40 shopper queries (10 per persona × 4 personas). Run weekly across ChatGPT, Perplexity, Claude, Gemini. Log citation share — do you appear, where — not click count.
- Schema audit — Rich Results Test or the GraphAPI on every template. Flag missing Product.offers, missing AggregateRating, broken Organization, wrong sameAs, unresolved @id chains.
- FAQ coverage map — open the top 20 category and product pages. Ask ChatGPT three shopper questions per page. Which does the AI get right from you, which from a competitor, which does nobody answer?
- Competitor citation-share benchmark — run the same 40-prompt panel against your three closest competitors. You need to know who's being cited instead of you before you plan what to take.
Phase 2 · Days 31–60 · Fix
Fix is engineering-heavy. The audit told you where the holes are; this phase closes them in priority order. The trap is “fix everything on every page at once.” Shopify stores have long tails — fixing the tail before the head wastes weeks. Treat this phase as Pareto work.
The order of operations
- Top 10% of SKUs first (days 31-40) — complete Product schema, clean Offer.priceSpecification, populated aggregateRating where available, canonical product IDs, clean gtin or mpn.
- Publish /llms.txt and /llms-full.txt — a concise crawler-friendly map of what your store is, what you sell, your policies, which pages are canonical. Not a replacement for HTML, but a beacon.
- Rewrite PDP copy for quotable claims — one sentence per spec, declarative, no fluff. Assistants quote short claims with clear antecedents. 'Water-resistant to 100 m' beats 'Our proprietary WaterArmor™ technology.'
- FAQPage schema on category + product pages — three to five questions per page, real shopper questions, 40-80 words each. Single biggest lever for category-query coverage.
- Merge thin category pages — collections with under six unique entities get rolled into the next layer up. Sparse pages hurt AI indexing; consolidation helps.
- Homepage entity copy — three paragraphs that name your brand, category, authority markers, and locations. Assistants stitch this into the brand graph.
Phase 3 · Days 61–90 · Amplify
By Phase 3 your on-site is trusted. The bottleneck becomes external: do other pages on the internet corroborate what your site says? Assistants cross-reference. Amplify is about earning those references honestly.
The five levers, in order of ROI
- Reddit + forum presence — participate in 3-4 subreddits relevant to your category. Answer shopper questions well, cite your product only when genuinely relevant. Do not shill. Reddit is overweighted by most assistants; authentic presence compounds.
- Niche reviewer outreach — pick 7-10 reviewers who already rank for your category queries on Google. Offer product, expect nothing. Roughly 30% convert into reviews within 90 days; those reviews become corroborating citations.
- Long-tail buying-guide cadence — publish two guides per week answering the questions your Phase 1 FAQ map flagged as uncovered. Earns direct citations and feeds the Google cross-check loop.
- Wikidata + entity corroboration — if your brand has any editorial coverage, a Wikidata entry with accurate sameAs, country, founding, and industry-code properties lifts entity recognition in every major assistant.
- Measure delta, double down — at Day 80 stop publishing for a week. Run a clean panel. Double budget on whatever moved; cut whatever didn't.
The weekly rituals that make it stick
The phases are the plan; the rituals are what actually ship. Without a weekly cadence GEO collapses back into a “project” that ends when someone moves on. Protect these five recurring slots on someone's calendar every week for the full 90 days and beyond.
- Monday — citation panel run (~60 min). The 40-query prompt set across ChatGPT · Perplexity · Claude · Gemini. Output: updated citation-share heatmap.
- Tuesday — log review (~45 min). Grep bot traffic; flag 4xx/5xx trends; confirm top pages are being fetched.
- Wednesday — gap-close content (~90 min). Ship one or two collection FAQs or buying guides that close a named citation gap.
- Thursday — schema regression (~30 min). Sample 20 PDPs through Rich Results Test; ticket any drift.
- Friday — competitor delta + outreach (~50 min). Who moved up, who dropped; outreach follow-ups for Reddit and reviewers.
What Day 90 looks like for stores that followed the blueprint
Across the 23 Shopify stores we tracked through the full 90 days, the median outcome at Day 90 was: citation share on panel prompts at 31% (vs 4% at Day 1), AI-cited server traffic at 2,800–3,200 weekly sessions (vs < 300), and one to three genuinely durable shoppers who arrived via a ChatGPT citation and placed an order. That last one is the number that matters. Everything else is instrumentation.