What is Product Truth?

Three questions an AI assistant asks

When someone asks an AI assistant for a product, it has to answer three different kinds of question about the company that sells it.

• Who is this company? Business Truth: name, locations, hours, lines carried, who to contact. It changes rarely.
• What does it sell? Product Truth: maker, part or model number, attributes with units, documents, what it replaces or fits. It changes when a catalog or data file changes.
• What can happen now? Operational Truth: which location has it, lead time, whether it can be quoted. It changes daily.

They fail differently

A wrong phone number is embarrassing. A wrong part number sends the wrong part. A stock level from last week promises something the branch can't deliver.

What makes a fact usable

A value on a web page isn't enough. For an AI assistant to state a product fact, the record needs:

• The source: the maker's data sheet, an industry data file, the company's own system, or a page Lehvel read.
• The date: when it was observed, and when someone with authority confirmed it.
• Who may see it: public, on request (a person answers), or internal.
• The unit and basis. "25 mm, ISO 15" is a fact. "25" isn't.

Each kind of fact has its own clock

A part number doesn't age. A spec sheet is good until the maker revises it. A listing or approval can lapse. Stock is stale by tomorrow.

Unknown is a real answer

The most important rule: missing information stays unknown. It never becomes "no".

If nobody has stated that two parts are interchangeable, the answer is "ask the branch", not "they don't fit". If the last stock reading is two days old, the answer is "call to confirm", not "in stock", and certainly not "out of stock".

That sounds cautious. It's what makes the "yes" answers trustworthy.

Stated, not guessed

Compatibility is where product information does the most damage when it's wrong: which part replaces a discontinued one, which brake pad fits which truck. Product Truth only states a relation that a source states, with that source attached. A replacement inferred from similar dimensions or a similar name is passed to a person to confirm.

Where it comes from

Product Truth isn't a new database a company has to fill in. It's read from the systems that already hold the answers:

When two sources disagree, both values are kept and the more authoritative one is fixed. Nothing is resolved by guessing.

• the catalog pages and structured data the company publishes;
• manufacturer data and industry standards, such as PIES, ETIM and ACES;
• spec sheets;
• an export from the ERP or DMS for stock and status.

What it's for

The same record feeds:

• the website's product pages and structured data;
• the answers AI assistants get when they ask;
• a check of what ChatGPT, Claude, Gemini and Perplexity say against it;
• one controlled action, a quote or inquiry request that a person answers.

Questions