For people whose name goes on the number

Be confident in the numbers your AI gives you.

SaysData sits between your AI and the official statistics it quotes. It searches the agencies’ own tables, tells your AI how each table is built, checks every query for mistakes AI commonly makes, passes on the agency’s caveats — and gives you a link to check how it got each result.

Connect SaysData to ClaudeSee the five safeguards

Data from Statistics Canada, Eurostat, and OECD. Works in Claude desktop and Claude on the web — no API keys.

Between your question and the number

Five places an answer can go wrong. A safeguard at each.

  1. 1FindThe agency’s own table, not a news story or a guess
  2. 2UnderstandUnits, scale, totals and missing cells, worked out first
  3. 3QueryKnown defects refused, known risks flagged
  4. 4InterpretThe agency’s footnotes and flags travel with the result
  5. 5VerifyA link to how it got every result

Your AI sounds just as sure when it’s wrong.

With official statistics, wrong rarely looks wrong. A figure in thousands read as dollars. A national total added to its own provinces. A series that changed definition halfway. The number is plausible, the prose is polished — and it’s your name on the report.

“…a model’s confidence is no guide to its reliability.”
Full Fact, “Wrong on Sunday, right on Monday”, its first AI benchmark report (2026)

Defence in depth

Five places an AI’s number goes wrong. A safeguard at each one.

No single check makes an answer trustworthy. SaysData stacks five, from finding the data to checking the result — each shown with real SaysData output.

01Find

The risk

It quotes a news story, a partial series — or a number it made up.

Asked for a statistic, a chatbot answers from whatever it has seen: secondary reports, an old release, a half-remembered figure. Sometimes it invents one that sounds right.

What SaysData does

It works from the agency’s own table.

Plain-language search over thousands of tables from Statistics Canada, Eurostat, and OECD. Each candidate says what it measures, where, over what period and how it’s broken down, so your AI can weigh them. The data is downloaded from the agency itself.

Evidence · search_datasets

search_datasets"wholesale sales by province excluding petroleum and oilseed and grain seasonally adjusted"…

Top-ranked candidate

Wholesale trade, sales20100074

Wholesale trade, sales in Retail and wholesale, Wholesale sales and inventories, from Monthly Wholesale Trade Survey. Data are monthly, with unadjusted and seasonally adjusted from 1993-01 to 2026-08. Geography: 14 areas at country (1, e.g. Canada); province or territory (13, e.g. Newfoundland and Labrador). …

02Understand

The risk

It doesn’t know how the table is built.

Thousands or millions? Which rows are totals? Is an empty cell zero, or suppressed? These live in the structure of the table, and an AI that skips the homework gets them wrong without saying so.

What SaysData does

Every table is profiled before your AI queries it.

SaysData profiles each file before serving it — units and scale, the period it covers, missing and suppressed cells — and checks it against the agency’s published structure; totals come from that structure or are recognised by their labels. Your AI can read that profile before it writes a query. A table that can’t be profiled isn’t served.

Evidence · describe_dataset

VALUE SCALE: every value is scaled "thousands": report it in that scale, or multiply by 1,000 to report base units — not both. …

template: pinned to "1" = Canada (the only member with nothing above it, with the others under it)

03Query

The risk

Its query is subtly wrong — and still returns a number.

Figures in units and in millions added as one. A national total added to its own provinces. Annual figures set against quarterly. The query runs, the result looks plausible, and nothing tells you.

What SaysData does

Queries with a known defect are refused. Risky ones come back flagged.

Every query has to say exactly which part of the table it wants. Automated checks refuse the defects they look for before anything runs — mixed units, scales or currencies the query doesn’t say how to keep apart, a breakdown the query never says how to handle, rows made up from typed-in numbers rather than read from the data — and return the fix. Queries with a risk they look for run, and come back with a warning.

Evidence · execute_sql

Refused — nothing ran

A query for Niger’s tax revenue that held rows at two scales — units and millions:

- slice "ner", UNIT_MULT [mixed-units]: slice ner holds 2 values in UNIT_MULT ("0", "6") and UNIT_MULT is neither pinned nor set to varying Fix: put UNIT_MULT in the slice's attributes: pin one value ({"values": ["0"]}) or set it to {"varying": true} and keep each value as its own series

Ran — with a warning

Wholesale sales for Canada and every province and territory in one query:

slice ws, "GEO_ID": "Canada" (1) is held beside members it contains. Summed with them, those members count more than once; ranked with them, it ranks among its own parts.

04Interpret

The risk

It reads the number without the fine print.

Breaks in a series, quality grades, confidentiality suppressions, a definition that changed. The agency publishes them as footnotes — the first thing an AI summary drops.

What SaysData does

The agency’s caveats travel with the result.

Results come back with their units and scale, the agency’s quality flags and footnotes, the file’s own date where it carries one, and the citation its licence asks for, dated the day SaysData downloaded the file. Each table’s profile links to the agency’s own documentation where it provides it.

Evidence · execute_sql notes and citation

A footnote on the table

Statistics Canada footnote 7, on this table: The Wholesale trade [41] series from 1993 to 2022 exclude data for oilseed and grain merchant wholesalers (NAICS 41112), and petroleum, … Data for oilseed and grain merchant wholesalers were included from January 2023, …

Quality grades

Rows whose "STATUS" is "A" are marked by Statistics Canada as: data quality: excellent.

Rows whose "STATUS" is "B" are marked by Statistics Canada as: data quality: very good.

Rows whose "STATUS" is "C" are marked by Statistics Canada as: data quality: good.

Units and citation

Every row of this slice is in Dollars, scale thousands (×1,000).

Adapted from Statistics Canada, Table 20-10-0074-01 Wholesale trade, sales, 1993-01 to 2026-08 (accessed on 2026-10-25). …

05Verify

The safeguard that lets you check the other four.

The risk

You can’t see how it got the number.

The answer arrives finished and confident, with a long list of websites and tool calls behind it. Checking it means redoing the work by hand — so too often it doesn’t get checked.

What SaysData does

Every result links to how your AI got it.

Each query result carries a link to its run: which table and which part of it, the exact query, the rows that came back, every warning and note your AI was given, and the citation — with the codes labelled and the raw request and response one click away. Open it beside the answer and check the number yourself. A link opens only for you, and every run is also in your History.

Evidence · the run page

saysdata.com/history/0e85690b-f600-4595-8a7d-f3695659746a

execute_sql

Answered
ReadableRaw request & response

Data used

The data may have been revised since.

Wholesale trade, sales (statcan 20100074)

Adapted from Statistics Canada, Table 20-10-0074-01 Wholesale trade, sales, …https://doi.org/10.25318/2010007401-eng

Query

SELECT "GEO", "REF_DATE", "VALUE", "STATUS" FROM ws ORDER BY "VALUE" DESC NULLS LAST

ws — the rows the query reads

From Wholesale trade, sales (statcan 20100074)

Column Code Means
GEO_IDGeographyevery value1Canada
2Newfoundland and Labrador
3Prince Edward Island
…
REF_DATEonly "2026-08"
Adjustments_IDAdjustmentsonly "2"2Seasonally adjusted
North_American_Industry_Classification_System_(NAICS)_IDNorth American Industry Classification System (NAICS)only "46"46Wholesale trade (excluding petroleum, …

Slice details 14 rows · 14 series · 1 period with a value

Result

14 rows · 4 columns

GEOREF_DATEVALUESTATUS
Canada2026-0891582303B
Ontario2026-0847791060A
Quebec2026-0815548431B
…

Warnings and notes

iAllWarnings (1)SaysData notes (4)Provider footnotes (11)

Slice ws

Warning

Aggregate or overlapping membersslice ws dimension GEO_ID Geography

… "Canada" (1) is held beside members it contains. …

SaysData note

Every row of this slice is in Dollars, scale thousands (×1,000).

Wholesale trade, sales (statcan 20100074)

SaysData note

Rows whose "STATUS" is "A" are marked by Statistics Canada as: data quality: excellent.

SaysData note

Rows whose "STATUS" is "B" are marked by Statistics Canada as: data quality: very good.

SaysData note

Rows whose "STATUS" is "C" are marked by Statistics Canada as: data quality: good.

On the whole table (10)

Provider footnote

Statistics Canada footnote 1, on this table: The total for farm product merchant wholesalers …

…

On a member above or below one this run selected (1)

Provider footnote

Statistics Canada footnote 6, on "Cannabis merchant wholesalers" (it is under the slice's "Wholesale trade (excluding petroleum, …"): Data for this Canadian industry are presented in unadjusted form …

Illustration of the run page for a real call made on 19 October 2026, with the notes, receipt and warning the same call returned on 28 October 2026; cuts marked “…”.

Your AI writes the answer. SaysData gives it the evidence.

We don’t claim to make an AI infallible, and nothing can. What SaysData does is make the official numbers easier to find, harder to get wrong and easier to read correctly — and give every result a run you can check. Confidence that comes from evidence you can see, not from how sure the answer sounds.

Built for people whose name goes on the number.

Policy analysts and researchers

Hand the quantitative groundwork to AI without staking a report on a guess.

Journalists

Check a figure against the agency’s own table, with its citation, before deadline.

Consultants

Know which table and query a figure came from before the client asks.

Public servants

Public data only, sources you can check, and a record of each query in your History.

Set up in three steps.

  1. 01

    Create your account

    Sign up, then open Getting Started for your connection address.

  2. 02

    Connect Claude

    Add SaysData to Claude with that address and sign in when asked. No API keys.

  3. 03

    Ask in your own words

    Ask about official statistics. Results come back with their units, notes and citation, and a link to the run.

Sources

Published by the agencies. Downloaded from their own releases.

Thousands of datasets. Each is profiled before it is queried, with its units, its licence citation, and the agency’s footnotes and flags where it publishes them.

Statistics CanadaEurostatOECD

Their data is used under each agency’s open-data terms. SaysData is independent and is not endorsed by or affiliated with any of them. Data sources and licences

Let your AI run the numbers — without relying on hope.

Sign up, add the connector, and ask your first question.

Get startedHow it worksRead the docs