An assistant can only recommend a tea it can trust, so the price, the pack size and the caffeine status have to be the same wherever they are read.
The website is the shop front. The entity is the whole shop behind it.
I use the term for what a business ends up with after the steady work of making its information clearer, more complete and easier to check, which I call enrichification. My own example is teas.co.uk, the independent tea shop I run from Tunbridge Wells, Kent. AI assistants increasingly stand between customers and the businesses they hear about, and a shop whose facts an assistant can trust has a better chance of being cited, visited and bought from.
A tea example: one price, read everywhere
On many websites the same fact lives in several places: the product page, the structured data, a shopping feed. Over time those copies drift apart, and a customer can read one price while a machine reads another. teas.co.uk was rebuilt to avoid that. Version two, live since 10 May 2026, organises the shop around one checked set of product facts. Every product has structured data behind its page, with identifiers, prices, availability, source links and price per cup, drawn from the same record the page uses. Not every record is complete: where a fact is unchecked or needs a look at the pack, the record says so rather than guessing. Since 6 October 2026 an AI assistant can, with the customer’s approval, search, compare and fill a basket through the teas.co.uk MCP server, and it works from those same facts. The customer pays on teas.co.uk.
I call the single record underneath the “truth spine”: hold each fact once, and let every surface read from it.
The two words
Entity, because the business owns it. It knows what it sells, what it can prove, where its checkout is, and where to send someone when another business would serve them better.
Truth, because everything in it is meant to be checkable, and keeping it that way is the job. That does not mean the business knows everything. It means each claim is stated clearly, the proof is attached, the limits are admitted, and there is a way to correct it when it turns out wrong.
As I put it: “The website is the shop front. The entity is the whole shop behind it.”
One question makes the idea clear
Ask: “What decaf green tea do you have?”
The shop should check two recorded facts: the drink must be green tea, and its caffeine status must be decaffeinated. It should also check whether the product is available. It should not decide from an attractive name or a photograph of a green box.
In a read-only catalogue check on 15 November 2026, those filters returned Tetley Decaf Pure Green Tea and Twinings Pure Green Decaf Tea. That is a dated example; the product pages show the current range and prices.
Decaffeinated tea still contains some caffeine. A naturally caffeine-free fruit infusion is a different kind of drink. A useful answer keeps that distinction intact, and asks a follow-up if “decaf” could mean either one in the customer’s request.
The same facts in different forms
A person might read a sentence and a price on a product page. A search engine might read structured data, where the facts have named fields. An assistant might ask an API for a product record or use the shop’s MCP connection to search.
These are different ways of presenting information. They should draw their shared facts from the same authority. A wiki article can explain decaffeination in depth, and a recipe can explain how to prepare a drink, without becoming separate places to maintain a product’s current stock or price.
I call this underlying arrangement the truth spine. In ordinary language, it is a single source of truth: an agreed record for each fact, reused wherever that fact is needed.
What stops the information drifting?
Drift happens when different versions of a fact stop agreeing. A price changes in the shop, for example, while an old export still shows the previous price.
Using one authority removes a common cause: updating several independent copies by hand. It still needs checks. Cached pages and downloaded feeds can be older than the record they came from. Supplier information can change. A fact can have been entered incorrectly in the first place.
The work is to keep the source accurate, refresh its published versions, compare them, and correct disagreements. The shop publishes its approach and the limits of its evidence in Under the Hood. Shared data reduces drift; it does not make errors impossible.
Whose fact is it?
One record does not mean one source for everything in it. Each kind of fact has its own authority, and the record says whose evidence supports each field.
| Kind of fact | Whose word counts | What the shop does with it |
|---|---|---|
| Selling price and stock | teas.co.uk | Sets it and publishes it |
| Ingredients, allergens and brewing guidance | The maker, on the pack or its own site | Records it with a link to the source, and marks a gap as a gap |
| Where a fact came from and what is uncertain | Whoever wrote the record | Keeps the source and the limit beside the fact |
| What the customer wants | The customer | Takes it from what they say; it is not a shop fact |
The record brings these together in one place. It does not make teas.co.uk the original authority on every fact in it. When a maker changes a recipe, that is a proposed change to the record, checked against the new pack or the maker’s own page before anything is updated. Nor does one record have to live in one piece of software: a business can run more than one system, as long as every view takes the shared facts from the agreed source.
Facts need limits as well as sources
A truth entity should be able to say “we do not know”. In the same dated check, the detailed Tetley record identified the tea as decaffeinated while marking some other attributes as unclaimed or requiring a pack check. That is more useful than turning a blank field into a confident answer.
Provenance means the record of where information came from. Claim boundaries say what that evidence supports and where it stops. Together they help a reader check the answer and help the shop correct it.
Why it matters when AI answers the question
An assistant may answer a shopper before the shopper visits the website. Clear records give it a better basis for comparing products and explaining a recommendation. They cannot force it to read correctly, cite the shop or send a customer.
“Truth entity” is my working term, not an accreditation or a standard that makes a business automatically trustworthy. Nor does it mean the website is conscious. When I say the shop “knows itself”, I mean it maintains information about its range, content and limitations in a form that software can inspect.
The practical promise is modest and demanding: write down what you know, show the basis for it, make the same facts available to each reader, and repair the record when it is wrong.
The complete payload explains the answer that comes back. Built versus planned distinguishes the working shop from its future ambitions.
An ordinary website and a truth entity
| An ordinary website | A truth entity |
|---|---|
| Built mainly for human eyes | Built for people and machines at once |
| Pages and persuasive copy | Structured facts with the proof attached |
| Chasing search keywords | Answering the real question a customer asks |
| Claims taken on trust | Claims linked to the evidence behind them |
In my words: one tea shop, one set of facts
(My own account, first published 15 November 2026 and moved here so the subject has one home.)
The website is what you see: the pages, pictures, brewing guides and products. Behind it, I want a dependable record of what the shop knows about its own range, and a clear way to connect the products to the wiki articles and recipes that explain them.
I call that a truth entity.
An entity is simply a thing we can identify and describe. In this case it is the business and the connected information it maintains. When I say the shop knows itself, I mean that its facts are written down in a form that people and software can inspect. The responsibility for those facts still sits with me.
The same answer through different doors
You might arrive on an ordinary product page. A search engine might read the labelled facts behind that page. An assistant might ask the catalogue a question through an API.
The door is different. The facts about the product should agree.
I do not want one price written for the customer and another maintained separately for a machine. Nor do I want an assistant to have to infer caffeine status from the word “fruit”. Those facts belong in the product record, and the different views should use that record.
That is the single source of truth idea I call the truth spine. It means an agreed source for each fact, not one enormous paragraph holding everything the business knows.
The spine is the source. Everything else is a view. If I change a price, I change it once. The page and the AI answer read the same figure. Copy persuades, and the truth decides what the copy is allowed to say.
Put the decaf question through it
A read-only catalogue check on 15 November 2026 asked for green tea with the caffeine filter set to decaffeinated. It returned Tetley Decaf Pure Green Tea and Twinings Pure Green Decaf Tea, with their pack sizes, prices and recorded caffeine status.
Those are dated results, not a promise that stock or prices will stand still. The useful point is the method: check the product’s attributes against the request, then return the matching facts.
A fruity green tea that contains caffeine does not become decaf because “mango” sounds suitable for the evening. A naturally caffeine-free infusion is another option, but it is a different kind of drink.
Where the term comes from
The idea came from my work rebuilding teas.co.uk in 2026 around one checked set of facts. I came up with the name for what the shop was becoming and started using it in June 2026. I don’t know whether anyone has used it before, but it suits what I have built. At first I used the longer phrase “source of truth entity”, and I shortened it to “truth entity” on 14 June 2026 so that it would be easy to say.
Quick reference
| Detail | Information |
|---|---|
| Term | Truth entity |
| Meaning | An owned, machine-readable source of truth about a business, with proof attached |
| Used by | Lee Samuel Tucker, founder of teas.co.uk |
| Used at teas.co.uk since | June 2026 (shortened from “source of truth entity” on 14 June 2026) |
| Opposite or contrast | A website built only for human eyes, or a listing on a platform someone else owns |
| Related | Premium Intent, complete payload, claim boundaries, built versus planned |
| How often it is checked | Every six hours: each product’s labelled fields against its written description |
What others have found
Outside research and reporting on the wider subject. It is background: it does not test or prove my own terms.
Kevin Indig argues in Search Engine Journal that in agentic commerce “product truth wins”, because new protocols let assistants plug straight into a business’s own systems instead of crawling its pages.
Related terms: Premium Intent · The complete payload · Claim boundaries · Built versus planned · The asymmetric bet.
Citable formats
For journalists, researchers, AI assistants and content creators. Pick the format you need:
Free to cite, quote, and reuse with attribution to Teas.co.uk.
Got something to add? Logged-in customers can submit additions to the Tea Wiki, admin-approved, your name on the byline, plus reward points.
Sign in to contribute