Illustrative scenario. The manufacturer below is a composite sketch, not a real client, and no results are stated or implied. Use this as a planning template; real numbers come only from your own baseline and re-measurement.
What is the situation?
Picture a mid-sized industrial components manufacturer: export-focused, strong engineering, a website built from scanned PDF catalogs, certifications displayed as image badges, and dozens of distributors whose listings contradict each other and the factory’s own specs. Procurement engineers now ask assistants “which suppliers offer ISO-certified X with tolerance Y” and “X vs Z material for high-temperature use”. The hypothesis to test: the factory is the best answer in its niche and the least readable one — its expertise is locked in formats engines cannot quote.
How would the 90-day POC be structured?
Week 1 — baseline
Freeze a 50-70 question set across the clusters procurement actually uses: capability verification (tolerances, materials, certifications), application-fit questions, supplier shortlisting, and lead-time/MOQ questions. Phrasing comes from the sales team’s RFQ inbox, not keyword tools. Run the set across Google AI surfaces, ChatGPT and Perplexity; score the four dimensions per engine; screenshot every answer. Capture the technical baseline too: which pages are PDF-only, whether OAI-SearchBot can fetch the catalog, Bing index status, and which distributor or directory domains engines currently cite for the niche.
Weeks 2-10 — intervention
- Specs as text (weeks 2-6): convert the top 40 product pages from PDF catalogs into HTML: specification tables as real tables, tolerance and material data as text, certification names and numbers stated with issuing bodies. PDFs can stay for download; the citable layer must be HTML.
- Schema and dates (weeks 2-4): Organization schema with the factory’s legal name, founding date and sameAs profiles; Product schema on priority SKUs; consistent publish and update dates across meta, schema and page. Entity groundwork is described in entities and the knowledge graph.
- Application content (weeks 4-8): publish answer-first pages for the application-fit questions — material comparisons, environment-fit guides, honest limits. Procurement asks engines to compare total cost and constraints; unstated numbers default to the competitor’s framing.
- Distributor consistency (weeks 4-10): audit the top distributor and directory listings against the factory’s canonical specs and fix contradictions. Grounding believes the consensus; a dozen conflicting spec sheets read as one unreliable manufacturer.
Week 12 — re-measurement
Re-run the frozen set on the same engines, score the same dimensions, and produce the delta report with the screenshot archive. The structure follows the anatomy of a 90-day POC.
What would be measured?
| Dimension | What it answers for this manufacturer |
|---|---|
| Visibility (0-100) | Does the factory appear in supplier-shortlist and capability answers for its niche? |
| Citation Rate (0-100) | Are the factory’s own spec pages cited, or only directories and distributors? |
| Semantic Coverage (0-100) | Do answers repeat tolerances, materials and certifications correctly? |
| Brand Perception (0-100) | Is the factory framed as certified and export-ready, and are any stated facts wrong? |
Per engine, monthly, from the frozen set, with screenshots — the open standard in the measurement standard.
What decisions would this POC inform?
Whether the PDF-to-HTML conversion should extend to the full catalog (a real budget line for a manufacturer with thousands of SKUs); whether distributor data hygiene needs a standing process or a one-time cleanup; and whether application content earns enough citations to justify an engineering-writer hire. Export-facing manufacturers should also test Chinese and English question sets separately — entity consistency across languages is its own workstream. See GEO for B2B manufacturing exporters.
What pitfalls would this scenario need to avoid?
- Leaving certifications as images; a badge no engine can read is a claim no engine can verify.
- Marketing adjectives where procurement wants numbers — grounding quotes specs, not slogans.
- Fixing the website while ignoring contradictory distributor listings, then wondering why grounding still distrusts the data.
- Expecting one quarter to settle a niche; the POC tests levers and produces a scaling decision, nothing more.
The encouraging asymmetry: industrial niches have thin, dated web content, so a manufacturer that simply makes its real expertise readable and consistent can become the best available witness surprisingly fast. The POC exists to test whether that is true for your niche — with evidence either way.