Illustrative scenario. The brand below is a composite sketch, not a real client, and no results are stated or implied. Use it as a planning template; the numbers a real POC produces belong to your own baseline and re-measurement.
What is the situation?
Picture a mid-sized fashion export brand: own-label womenswear, selling through its own site and two marketplaces into North America and Europe, strong product photography, thin text content, no structured data, and a review profile concentrated on one marketplace. Shoppers increasingly ask assistants “best linen dresses for summer weddings” or “is this brand true to size” — questions the brand has never checked. The hypothesis to test: the brand is invisible in AI answers for its own category, and its marketplace listings are the only witnesses engines can find.
How would the 90-day POC be structured?
Week 1 — baseline
Freeze a question set of 60-80 real shopping questions across three clusters: category best-of (linen dresses, office-to-evening), fit-and-quality verification (sizing, fabric, returns), and brand-direct questions. Collect the phrasing from customer support logs and marketplace Q&A, not from keyword tools. Run the set across Google AI surfaces, ChatGPT and Perplexity, score the four dimensions per engine, and screenshot every answer. Also capture the technical baseline: crawler access, Bing index status, schema coverage, and which third-party domains engines currently cite in the category.
Weeks 2-10 — intervention
- Technical (weeks 2-4): verify Bing Webmaster Tools, enable IndexNow, confirm OAI-SearchBot and PerplexityBot access, deploy Organization, Product, Offer and FAQPage schema with dates consistent across meta, schema and page.
- Content (weeks 3-8): rewrite the top 30 category and product pages answer-first: sizing guidance as text, fabric and care specs as crawlable tables, honest “who this is for” blocks, and comparison content for the category’s real versus-questions. See comparison content.
- Source graph (weeks 4-10): claim and complete profiles on the review platforms the baseline citation list revealed, start a compliant review invitation flow, and pitch two data-backed stories (sizing transparency, fabric sourcing) to the editorial outlets engines already cite. See source graph building.
Week 12 — re-measurement
Re-run the identical frozen question set on the same engines, score the same four dimensions, and compare deltas against the week-1 archive. The deliverable is the delta report plus the screenshot evidence, not a promise. The full method is described in the anatomy of a 90-day POC.
What would be measured?
| Dimension | What it answers for this brand |
|---|---|
| Visibility (0-100) | Does the brand or its products appear in answers for the category clusters at all? |
| Citation Rate (0-100) | Are the brand’s own pages cited, or do engines rely on marketplaces and third parties? |
| Semantic Coverage (0-100) | Which attributes — sizing, fabric, price tier, shipping — do answers correctly repeat? |
| Brand Perception (0-100) | What do answers say about quality, fit and returns, and is any of it wrong? |
Each dimension is scored per engine from the frozen set, monthly, with screenshots as the audit trail — the open standard described in the measurement standard.
What decisions would the POC inform?
Three: whether to scale content engineering to the full catalog or keep it on hero categories; whether marketplace presence is helping or cannibalizing owned-site citations; and whether the source-graph work (reviews, editorial) justifies an always-on PR retainer. A well-run POC ends with a decision memo, not a victory lap — some deltas will be flat, and knowing which interventions did nothing is worth the 90 days by itself.
What pitfalls would this scenario need to avoid?
- Changing the question set mid-flight, which destroys comparability.
- Judging success on referral sessions while the category is majority zero-click.
- Seeding fake reviews to move Brand Perception — detectable, penalized, and fatal to the evidence’s credibility.
- Claiming POC deltas as projected revenue; the POC measures visibility mechanics, and revenue modeling is a separate, explicitly labeled exercise.
Run honestly, even a flat result is a win: the brand learns exactly where it stands in the answer layer, which levers connect to movement, and what a scaled program would cost per lever. That is what a proof of concept is for.