WhatDrink editorial policy

• • Decision usefulness comes before word count.
• • Evidence comes before claims.
• • Trade-offs are stated clearly.
• • Wording is conservative where certainty is low.
• • Limitations are visible, not hidden.
• • We do not manufacture freshness with fake dates.
• • We do not publish unreviewed recommendations.
• • We do not make unsupported health, allergy, or medical claims.
• • We do not dump raw taxonomy lists as if they were content.
• • We do not add claims purely to make a page longer.

The rule behind every specific claim on the site is the same one we describe in our methodology:

• • We derive facts through calculations over real, structured data.
• • We extract facts from identifiable source evidence.
• • We use generated text only for supported interpretation and wording.
• • We never fabricate missing specifics.
• • We omit the claim when the evidence is insufficient.

An honest gap is better than a confident guess. AI-assisted wording helps present what the evidence already supports; it is never itself treated as a source of fact.

Content is produced by a combination of code, approved data, and AI-assisted editing, each with a defined role:

• • Code handles deterministic checks and calculations.
• • Approved data contracts constrain what public output can say.
• • AI may assist with organisation, classification support, summarization, and wording.
• • Taxonomy values must match approved entities.
• • Weak or high-risk claims are suppressed.
• • Candidate recommendations are not treated as verified facts until they are reviewed and promoted.

For the mechanics of this pipeline, see our methodology.

Recommendations only appear when we can justify them. Our rules:

• • Similarity must have a visible reason.
• • Substitution guidance must explain what changes.
• • Pairings must be supported by evidence.
• • Recommendations must identify a useful trade-off.
• • Weak or unreviewed relationships stay unpublished.

Any sponsored placement or commercial relationship that affects displayed content will be identified clearly. Where advertising or promoted placements appear, they are presented so you can tell them apart from editorial content and data-driven guidance.

We would rather fix an error than defend it. If something on the site is wrong, please tell us on the corrections page. Material changes we make are recorded on data updates.