AI tools

The Posts module bundles several AI-assisted features that help editorial teams work faster without sacrificing quality — automatic category suggestion for new articles, topic suggestions for what to write next, fact-checking on saved content, image generation from a reference-image library, and the automatic translation pipeline that keeps every locale in sync.

Last updated 2 August 2026postsaiautomation

This article gives an inventory of the AI-assisted surface, when each feature helps, when it hurts, and what to consider when enabling them for your editorial workflow. Deeper coverage of the biggest one — translation — lives in Translations; the image-generation flow is in Gallery and images.

Automatic category suggestion

Every post editor has a Suggest category button next to the main-category dropdown. Click it and the platform:

Reads the current article title and body.
Sends both to the AI along with a list of your active categories and their descriptions.
Returns the single best-matching category.
Displays the suggestion — one click accepts and populates the main-category field.

Practical notes:

Only active categories are candidates. Deactivated categories are excluded so a suggestion never nominates a section you have deliberately retired.
Category descriptions steer the model. The description you write for each category is what the AI reads to understand what "belongs" there. Vague descriptions produce vague suggestions; concrete descriptions with keywords produce reliable ones. See Post categories for best practices on writing category descriptions.
The suggestion is not automatic. Nothing is set until you click accept. Suggestions are always human-reviewed.
Suggestions are one-shot. There is no ongoing "watch this article and re-suggest if it changes"; the suggestion happens when you click, using the current article content at that moment.

Use for the first draft of a new article, or when reorganising an imported batch of unclassified content. Skip when you already know exactly where the article belongs.

Topic suggestions

The AI-suggested topics button in the top-right corner of /post (next to the Add button) opens a modal that suggests new article ideas your editorial team could pursue.

The suggestion flow:

The platform reads the current category tree, the topics of recent articles, and (optionally) any competitor sites you have configured as reference.
It analyses coverage gaps — categories that are under-served, topics competitors write about but you do not.
It returns a list of 5–15 concrete article-title suggestions, each paired with a target category, a suggested angle, and a short reasoning ("Category X has no article about Y since 2024; competitor A published a piece on this last week.").

Click a suggestion to open a pre-filled create-post modal with the title, category, and angle already populated. From there the flow is a normal post creation.

Suggestions are a discovery tool, not a mandate. They surface opportunities your editorial team can accept, reject, or ignore based on strategic judgement.

Fact-check

The Fact-check button in the post detail's header runs the AI fact-checker on the article's body content. It:

Extracts specific factual claims from the article — dates, numbers, quotes, names, causal statements.
For each claim, evaluates whether it is likely true, likely false, or unverifiable with confidence.
Returns a table of findings: category (error / falsehood / recommendation), severity (low / medium / high), the excerpt from the article that triggered the finding, the issue description, justification, and a suggested fix.

A clean article shows "no findings". An article with issues shows the table for editorial review; each finding is a suggestion, never an automatic edit.

Practical caveats:

The AI is not an oracle. Fact-checker output is a starting point. Every finding should be validated by a human before an edit is made.
Coverage varies by domain. Well-known facts about established topics (history, standard science) fact-check reliably. Niche technical claims, very recent events, and in-house proprietary information may produce false positives or false negatives.
Run before publishing, not after. Fact-check is a pre-publication safety net. Running it on an already-live article that turns out to have issues does not automatically correct them; you have to make the edits yourself.

Use for articles with a high risk of factual error — investigative pieces, technical explainers with specific numbers, news summaries. Skip for opinion pieces where the "facts" are subjective by design.

AI image generation

Covered in depth in Gallery and images. The short version: the gallery tab has a Generate with AI form that produces an image from a prompt, steered by the organisation-wide reference-image library. Generated images can be added to the gallery like any manually uploaded image.

The generation uses the mood prompt and active reference images from the organisation's library (managed at /settings/posts) to produce visually consistent output across every author's articles.

Automatic translation

The biggest AI surface in the Posts module — an automatic translation pipeline that keeps every enabled locale in sync with the primary-language source. Covered in full detail in Translations; the summary here:

Trigger — click the Auto-translate button in the post detail sidebar (or the row context menu) when target locales are missing or stale.
Model chain — every translation is attempted against three models from two providers in fallback order; the first successful validated response wins.
Validation — the platform rejects empty content when the source had content, extreme length mismatches, and source-echo responses. Failed validation retries the next model.
Atomicity — either a translation lands in full and is stamped as fresh, or nothing lands and the row stays in its previous state for the next attempt.
Freshness — the platform tracks which source version each translation was generated from and marks translations stale when the source is edited. See History and versioning for how source snapshots interact.

The Translations tab on /post is the operator's dashboard for translation status across the whole organisation, with one-click actions for bulk translate, refresh stale, and route repair.

Combining AI tools

The AI tools cover four discrete moments in the editorial flow:

Discovery — topic suggestions surface what to write about.
Classification — category suggestion places an article in the right section of the tree.
Fact validation — fact-check catches errors before publication.
Distribution — auto-translate makes the article available in every language.

Used together, they collapse work that would otherwise take a person hours into decisions the editor still owns but no longer originates from scratch.

What AI does NOT do

To set expectations honestly:

Write full articles. There is no "generate an article on topic X" button. The topic suggester proposes titles; execution stays with the writer.
Rewrite for tone. No "make this more casual / more formal / more concise" transformation is exposed in the editor. Voice and tone remain manual.
Moderate comments. Comment moderation is fully manual (see Comments and moderation); no AI spam-scoring is applied to visible comments.
Score reader feedback. The 1–5 star ratings are raw human input, unfiltered.
Replace a human editor. Everything AI-generated in the Posts module is a suggestion or a starting point that a human accepts, edits, or rejects. Nothing lands on the public site without a human clicking a button.

Cost considerations

Every AI call consumes credits from the organisation's AI budget:

Category suggestion — one call per click. Cheap.
Topic suggestions — one call per Suggest-topics modal open. Moderate cost (larger context including the full category tree).
Fact-check — one or a few calls per run, depending on article length. Moderate cost.
Image generation — one call per Generate click. Higher cost than text calls.
Auto-translate — one call per (article, target locale). At scale (hundreds of articles × several locales) this is the biggest driver.

Auto-translate is careful about cost — it skips locales already in sync (fresh), and the Translate missing action further narrows to only fully-missing locales. Both are visible in the Translations tab. When you have budget constraints, prefer Translate missing and hand-pick individual articles for full refresh.

Configuration

The AI features are enabled per-organisation. Check with your platform administrator whether all features are turned on. Some features (image generation, fact-check) may be tier-gated on certain subscription plans.

Configuration surfaces:

/settings/posts — the reference-image library and mood prompt for image generation.
Organisation AI settings — general AI provider and model preferences (usually managed by the platform administrator, not the editorial team).

Post management overview
Translations — the biggest AI feature.
Gallery and images — image generation detail.
Post categories — how category descriptions steer the classifier.
Post editor