Waterworks & Utility is being reranked by AI. Is your catalog readable?

A municipal contractor sourcing a 12-inch C900 PVC gasketed pipe for a water main replacement doesn't start on a distributor's site anymore. He asks ChatGPT or Perplexity to confirm the pressure class, DR rating, and gasket compatibility, then asks which supplier has it in stock. If your product data can't answer that question in a form the model can parse, it moves to whichever distributor's data can — and that distributor gets the quote request. Waterworks is a category where getting the spec wrong isn't a minor inconvenience, which makes it one of the categories where AI answer engines are least forgiving of thin data.

Buying research has already moved into the chat window

This shift isn't specific to waterworks, but waterworks sits squarely inside it. Recent industry reporting puts AI tool usage in B2B supplier research at 66%, with roughly 90% of those buyers trusting the recommendations the tools surface (Trax Technologies, 66% of B2B Buyers Now Use AI For Supplier Research). Gartner has separately projected that traditional search engine volume could fall roughly 25% by 2026 as research shifts to AI assistants (Redevolution, AI Search and B2B: What Engineering Companies Need to Know in 2026). For engineers and contractors specifying pipe, valves, hydrants, and meters, the shortlist of eligible suppliers is often decided before anyone opens a distributor's website.

The mechanism matters as much as the volume. Answer engines like Perplexity typically synthesize a response from a small set of cited sources — often two to seven per answer, not the ten blue links of a Google page (Bignewsnetwork, AI Search Has Created a New Visibility Problem for B2B Manufacturers). Getting cited is harder than ranking on page one used to be, and being left out isn't a lower position — it's no visibility at all. For a waterworks distributor competing against a handful of national players with deep catalogs, that's a real threat.

Why a full catalog can still look empty to a model

Most waterworks ERP exports were built to print a pick ticket, not to answer a spec question. A typical feed row looks like this:

Raw ERP feed description (as-is):

DI GATE VLV 8IN MJXMJ 250 CL EPOXY

That string works fine if you already know your ERP's abbreviation conventions. It tells a model almost nothing reliable, because the material, connection type, pressure class, and coating are compressed into one unstructured token with no labels attached. A language model — or a contractor typing the same question into a chat window — has no clean way to confirm this part matches a 250 psi working pressure requirement for a water main tie-in.

Enriched, the same SKU looks like this:

That's the gap between a string an ERP can print and a set of labeled attributes a model can reason over. Once pressure class, connection type, and compliance standard are explicit, discrete fields, an answer engine can match "AWWA-compliant 8-inch mechanical joint gate valve rated for a 250 psi water main" to the actual SKU instead of guessing at what "MJxMJ" or "250 CL" mean.

Ask an answer engine: what this looks like in practice

Here's a query a utility engineer or waterworks contractor could plausibly run today:

"I need an 8-inch ductile iron gate valve, mechanical joint both ends, Class 250, AWWA C509 compliant, epoxy-coated for potable water. Which distributors carry it and can deliver to a municipal job site this week?"

The engine is parsing that query for size, material, connection type, pressure class, standard compliance, and availability. If your product page or feed encodes those as clean, explicitly labeled attributes — in visible page content, in schema.org Product and additionalProperty markup, or in a feed a retrieval layer can actually parse — you're a candidate for the answer. If that information only lives in a scanned spec sheet PDF or a compressed ERP string, the model can't verify the match, and it will cite whichever competitor made the decision easy. Structured, explicitly labeled data is consistently reported as more likely to surface in AI-generated answers than unstructured pages carrying the same information, because the model doesn't have to infer what a field means.

What machine-readable actually requires

None of this is exotic for waterworks distributors — it's the same data cleanup most already know they're behind on, with a sharper reason it matters now:

• Split compressed spec strings into discrete, labeled fields — material, connection type, pressure class, coating, standard compliance.
• Standardize abbreviations and units so MJ, PE, FLG, and pressure classes aren't ambiguous or missing across manufacturer lines.
• Fill the gaps supplier catalogs routinely leave blank — AWWA/ASTM standard citations and pressure ratings are among the fields most often dropped in raw feeds.
• Keep values current as manufacturers revise specs or supersede parts, so an answer engine isn't citing a discontinued fitting.

Done by hand, this enrichment runs somewhere in the range of 30-45 minutes per SKU — pulling supplier docs, normalizing abbreviations, filling gaps, re-verifying. Across tens of thousands of active waterworks SKUs spanning multiple manufacturer lines, that's not a project a data team finishes before the next refresh makes it stale again. The values still have to come from somewhere real — extracted and quality-scored against supplier documentation, not invented, which matters more in a category where a wrong pressure rating is a safety issue.

Where this fits for distributors

Your ERP or PIM stays the system of record — Anglera doesn't replace it and has nothing to do with your CRM. It sits on top of whatever you already run (Akeneo, Salsify, inriver, Stibo, Syndigo, Pimcore, Informatica, or nothing formal — a flat export is enough to start) and continuously extracts, scores, and gap-fills the attributes that turn a compressed spec string into something an answer engine can match against a real query. Most distributors can get a meaningful slice of a waterworks catalog to that state within 30 days, not a multi-year systems project. Getting there first isn't about chasing better rankings — it's making sure your SKU is the one an AI answer engine can confidently recommend when a buyer asks the question that used to start with a call to the counter.