Guide

AI-Assisted Workflows (Pro)

23 min read 9 views

Version

1.0

Last Version Update

October 4, 2026

Effective From

October 4, 2026

Review Cycle

Annual

Next Review Due

October 2027

AI-Assisted Workflows (Pro)

These features require the Dantian Policy Management Pro add-on. AI features are always opt-in and require you to provide your own API key from OpenAI or Anthropic. The plugin never uses a shared or plugin-vendor AI account. AI features also need an active Pro licence.

The Pro add-on includes optional AI automation for content workflows. One master switch turns AI on or off, and most features have their own setting as well, so you can use as many or as few as you want. Documents classified Confidential are never sent to an AI provider, under any setting.

Supported AI providers

OpenAI — GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5.4 nano and GPT-4.1
Anthropic Claude — Claude Sonnet 4.6 (the default and recommended model), Claude Opus 4.7 and Claude Haiku 4.5. Optional prompt caching reduces cost on repeated calls.

You don’t choose the provider separately — it is detected automatically from the model you pick. You can also choose cheaper models for individual features (search intent, related documents, interlinking, vision extraction and the framework tools).

You bring your own API key. Cost is billed directly by the provider at your account’s rates. The plugin keeps a local log of every AI call — the feature, model, token counts and estimated cost — so you can monitor spend (see AI Cost Report below).

Setting up AI

Go to Policy Management → Settings → AI & Automation.
Open AI API Configuration and enter your OpenAI API Key and/or Anthropic API Key — you only need a key for the provider whose models you use. Keys are stored encrypted. Once saved, the field shows Key on file: yes; leave it blank on later saves to keep the existing key. (For extra security, your developer can supply the key in the site’s configuration file instead, so it isn’t stored in the database.)
Choose the Default AI Model.
Under Usage & Limits, switch on Enable AI features — the master switch. While it is off, every AI feature stays inactive even if its own toggle is on.
Review the limits in the same section:
- Daily API Limit — maximum AI calls per day across the site (default 100; 10–10,000). The current day’s usage is shown.
- Per-User Daily API Limit — maximum calls one user can trigger per day (default 50; 0–10,000, where 0 = no per-user cap). Your own usage today is shown.
- Require Consent — show a confirmation dialog before FAQ generation explaining that document content will be sent to your provider (on by default).
- Prompt Caching — Anthropic prompt caching (on by default). Anthropic keeps the instructions part of each request for 5 minutes, so repeated calls in a batch cost up to 90% less for that part; the first call costs about 25% more. Turn it off if you mostly run one-off, single-document requests.
- Vision Extraction Model — the model used for AI Vision Extract on scanned PDFs (default Claude Sonnet 4.6). Choices: Claude Sonnet 4.6, Claude Opus 4.7, Claude Haiku 4.5, GPT-4o and GPT-5.4.
Review AI Content Privacy (see below).
Turn on the individual features you want in their sections, then save.

The daily counters reset at midnight UTC (late morning in eastern Australia), not at midnight on your site’s clock.

There is no “test connection” button — generate something small (for example, FAQs for one document) to confirm your key and model work.

The Features tab of Settings also shows an AI features row that tells you whether AI is on and links to Configure AI features →. It is a shortcut only; the real switch is on the AI & Automation tab.

What the master switch covers

Enable AI features turns on and off FAQ generation, document summaries, Smart Search intent, related-document suggestions, comprehension-question generation and the automatic question review. While it is off, the AI FAQ Generation box and the FAQs → AI Generated screen are hidden.

Three AI tools have their own controls and are not turned off by the master switch:

the AI Scan in Document Interlinking — controlled by AI-Assisted Interlinking;
Imported Drafts AI suggestions and AI Vision Extract — available whenever an API key is saved. Use the Skip AI analysis option on an import batch, or a blocked confidentiality level, to keep documents away from them.

The Regulatory Framework AI tools have their own settings, described in Regulatory Framework Mapping.

What you see without Pro

The Free plugin has no AI features and no AI & Automation settings tab. Free users don’t see any of the boxes, buttons or screens on this page. The Free plugin’s File Import shows a Skip AI analysis for this import batch checkbox, which only matters once Pro is active.

How AI requests run in the background

Most AI buttons don’t make you wait on a single long page load. When you click Generate FAQs, Generate Summary, Suggest Related Documents, Generate with AI (comprehension questions), AI Scan (interlinking), Run AI suggestions (Imported Drafts) or start a Vision extraction, this happens:

Your request joins the site’s AI queue and the button shows that it’s working.
WordPress’s built-in scheduler picks up the job straight away and sends it to your AI provider. Jobs run one at a time, oldest first, using your user account’s permissions.
Your browser checks for the result every 3 seconds, then less often for longer jobs. If there’s no result after 6 minutes, the page stops waiting and shows an error.
Just before the job is sent, the document’s confidentiality level is checked again. If someone marked it Confidential (or another blocked level) after you clicked, the job is refused rather than sent.

Finished and failed jobs are removed from the queue about an hour later. While AI features are switched on, an hourly background check also picks up any job that got stuck because the scheduler missed it.

The Generate FAQs (AI) bulk action on the documents list doesn’t use the queue. It runs while the page reloads.

AI FAQ Generation

Automatically generate FAQ content from a document’s text.

How it works:

Open a document in the WordPress editor.
In the AI FAQ Generation box in the sidebar, click Generate FAQs. (This box is shown to administrators when AI is enabled.)
If Require Consent is on, confirm that the document contains nothing that shouldn’t be sent to an external AI service.
The plugin sends the document content to your AI provider, which returns likely questions and answers. Existing FAQs are checked to avoid duplicates.
In Preview Generated FAQs, review the results. Anything similar to an existing FAQ is flagged and unticked by default; where a new answer adds to an existing FAQ, you can review a proposed merge and apply it.
Select the FAQs you want and click Create FAQs.

New FAQs are created as drafts for review unless you turn on Auto-publish FAQs. AI-created draft FAQs are easy to find under FAQs → AI Generated (see below).

Who can use it: the AI FAQ Generation box appears only for administrators, and only while Enable AI features is on. FAQ generation works on draft and published documents.

Generating for several documents: on the documents list, tick the documents, choose Generate FAQs (AI) from Bulk actions and click Apply. FAQs are created straight away (as drafts unless auto-publish is on), and a notice reports “AI FAQ Generation complete: X FAQs created, Y duplicates skipped, Z errors.” If the daily limit is reached part-way through, the remaining documents are skipped.

Summaries come along for free: if the document has no AI summary yet, or its summary is from an older version, generating FAQs also refreshes the summary. That is an extra AI call, which counts towards your limits.

FAQ Generation settings (on the AI & Automation tab):

FAQs per Document — how many FAQs to generate (1–20, default 5)
Auto-publish FAQs — publish generated FAQs immediately instead of creating drafts (off by default)
Dedup Catalogue Size — how many existing FAQ titles to send for duplicate checking (default 100; 0 = all)
Inherit Audience — give generated FAQs the same visibility restrictions as the source document (on by default)

What is sent to the AI: the document content and the active prompt, plus existing FAQ titles for duplicate checking. Documents at a blocked confidentiality level are refused (see Privacy and data handling).

The AI Generated FAQs screen

While Enable AI features is on, the FAQs menu gains an AI Generated item. It opens the FAQ list filtered to draft FAQs created by the AI, so you can review them in one place. A red badge on the menu item shows how many are waiting. The same list is available from the AI Generated (Review) link above the FAQ list.

The FAQ list also has an AI Confidence column (sortable). It shows the score the AI gave each generated answer: 90% and above is marked high, 70–89% medium, and lower scores low. FAQs you wrote yourself show “—”.

To review AI-generated FAQs:

Go to FAQs → AI Generated.
Sort by AI Confidence to check the lowest-scoring answers first.
Open each FAQ, check the answer against the source document, edit it if needed, then Publish it — or move it to the bin.

Anyone who can edit posts (editors and administrators) can see this screen. If you turn on Auto-publish FAQs, generated FAQs are published straight away and don’t appear here, because the screen lists drafts only.

AI Document Summaries

Generate a plain-language executive summary of a document. The summary is displayed above the document content on the document page — useful for long, technical procedures where staff benefit from a plain-English overview before reading the full content.

Generating a summary:

Open the document in the editor.
In the AI FAQ Generation box, click Generate Summary (or Regenerate Summary if one already exists).
Save the document.

If the document has moved to a newer version since the summary was generated, the box tells you so (“Summary is from v1.0 — document is now v1.1. Regenerate to update.”). Missing or outdated summaries are also refreshed automatically whenever you generate FAQs or suggest related documents for that document.

Summaries can only be generated for published documents. The AI writes a plain-text overview of two or three sentences covering the document’s purpose, scope and who it applies to.

Whether summaries appear on the front end is controlled by AI Document Summaries (Display AI-generated summaries on document pages, on by default). Each document can override this with its Display option: Use global setting, Always show or Always hide.

AI-Assisted Related Documents

Suggest related documents based on content similarity. The documents you save appear in the Related Documents section on the document page.

Getting AI suggestions:

Open the document in the editor.
In the AI FAQ Generation box, click Suggest Related Documents.
The plugin sends the current document’s content and a catalogue of your other documents to the AI.
The AI returns a list scored by relevance. High-relevance matches are shown by default; medium-relevance matches can be revealed with a toggle.
Choose the documents that are genuinely related and save them.

Each suggestion can carry an audience warning, for example “No audience overlap. This document is for Nurses; target is for Managers.” That means some readers of this document won’t be able to open the suggested one.

The saved list appears in the box under Saved related documents. Click × next to a document to remove it. To add a document yourself, type in Search documents to add… and click +. No AI is involved.

Suggestions are remembered for 24 hours, so clicking the button again shows the same list without another AI call. Hold Shift while you click Suggest Related Documents to fetch fresh suggestions. Only published documents can get suggestions, and only other published documents are considered.

Related Documents settings: Related Documents Model, Relevance Thresholds (high relevance minimum 80, medium relevance minimum 60 by default — anything below medium is discarded) and Max Suggestions (1–10, default 5).

AI Document Interlinking

Detect references to other documents within a document’s content and turn them into links.

Enabling interlinking: in Settings → AI & Automation → Document Interlinking, switch on Auto-Interlinking (Enable document interlinking). This works without AI by matching document titles. Also switch on AI-Assisted Interlinking to have the AI detect paraphrased and contextual references, and optionally choose an Interlinking Model.

Settings (all off by default):

Auto-Interlinking — turns on the Document Interlinks box, title-match scanning and links on the front end.
AI-Assisted Interlinking — adds the AI Scan button.
Interlinking Model — the model for AI Scan; leave it blank to use the Default AI Model.

How it works:

Open a document in the editor. Each time you save a published document, the plugin looks for the exact titles of your other documents in its text. Matches are listed in the Document Interlinks box under Pending Suggestions, marked Title match. Drafts aren’t scanned until they’re published.
Click Re-scan to refresh title matches at any time. If AI-Assisted Interlinking is on, click AI Scan to have the AI look for paraphrased references, such as “our privacy rules” for a document called Privacy Policy. The scan runs in the background (see How AI requests run in the background); you can Stop waiting for it. When it finishes, it reports how many new suggestions it added, marked AI-detected.
For each suggestion, use View in Context to see where the reference appears, then click Approve or Dismiss. Accept All Title Matches approves every title-match suggestion in one go.
Approving a suggestion inserts the link into the document text in the editor, at the first place the reference appears. Save the document to keep the change.
Approved links are listed under Approved Interlinks, with an AI badge on the ones the AI found. Click Remove to take a link out. Dismissed suggestions are kept under Dismissed, where you can Restore them.

On the document page, approved references are also linked when the page is displayed. A reader only sees a link if they’re allowed to view the target document, so restricted documents aren’t advertised to people who can’t open them.

Who can use it: anyone who can edit the document. While Auto-Interlinking is off, the box shows a short explanation and a link to Settings → AI & Automation instead.

What is sent to the AI (AI Scan only): the document’s content and a list of your other document titles. Title-match scanning happens on your own server and sends nothing. The AI Scan is refused for documents at a blocked confidentiality level, and each scan counts towards your daily limits.

Interlinking helps staff navigate between related policies and reduces the “I read this policy but didn’t know to look at the related procedure” problem.

AI Search Intent Rewriting

When enabled, question-like search queries are interpreted by the AI and turned into a short keyword phrase before searching. This means staff who search “what happens if I’m late to work” can find the attendance and punctuality policy, even if those exact words don’t appear in the document.

Enabling search intent:

In Settings → AI & Automation → Smart Search, tick AI Search Intent (Use AI to interpret natural-language search queries). It requires the master Enable AI features switch. Optionally choose a cheaper Search Intent Model.

It applies automatically to the plugin’s front-end search. Staff type their question in natural language and see the results; any extra results found through smart search are added below the keyword results, and if there are no exact keyword matches the page shows what it searched for instead. If the AI is unavailable, search falls back to standard keyword matching with no error shown. Searchers see a short notice that smart search sends their search text to your AI provider.

Short keyword searches are not sent to the AI. Rewrites are cached for 10 minutes and count towards your daily API limit.

What is sent: the search query text only. No document content is sent for search rewriting.

AI-Generated Comprehension Questions

With AI features enabled, the Pro add-on can draft comprehension questions for a document.

Open the document in the editor.
In the Acknowledgement Content & Comprehension box (shown when the document requires acknowledgement), find Generate with AI (Pro).
Choose how many Drafts to create (1–12, default 6) and click Generate with AI. The request runs in the background; the status line shows Generating… and then Drafts ready for review.
The AI reads the document and returns draft questions, choosing the type that suits each fact: Single choice, True / False or Multiple correct. The correct answers are marked ✓. It is told to test only facts stated in the document, to avoid “all of the above” and not to invent numbers.
Review each draft. Click Approve to move it into the live question pool, or Discard to delete it.
Save the document. Approved questions are added to the question list in the editor, but they’re only kept once you save.

Drafts are never published automatically, so you always review questions before staff see them. Drafts you haven’t approved or discarded stay on the document until you deal with them.

Who can use it: anyone who can edit the document, while Enable AI features is on.

What is sent to the AI: the document title and its text, with formatting removed. Only about the first 12,000 characters are sent, so for a very long document, questions come from the earlier sections. Documents at a blocked confidentiality level are refused.

Question review after a new version

When a minor or major new version of a document is approved, the AI automatically checks the document’s existing comprehension questions against the updated text. Patch versions don’t trigger a review. The review only runs when Enable AI features is on, the document uses comprehension questions and already has at least one, and its confidentiality level isn’t blocked.

The AI sorts each question into one of three groups:

Still valid — stays in the live pool, unchanged.
Needs review — taken out of the live pool and flagged for you.
Removal suggested — taken out of the live pool and flagged for removal.

It can also suggest new questions for substantially new content (New suggestion).

When anything is flagged, editors see a notice on the document’s edit screen, for example “2 comprehension questions need your review after the recent version update.” In the Question Review (Pro) panel inside the Acknowledgement Content & Comprehension box:

For each flagged question, click Restore to pool to put it back, or Remove to drop it.
For each new suggestion, click Approve to add it to the live pool, or Discard.
Save the document.

Flagged questions aren’t shown to staff while they wait for your decision, so staff never answer a question that no longer matches the policy.

AI Cost Report

A built-in cost report shows your AI API spend:

Go to Policy Management → Analytics and open the AI Costs tab (requires the Manage settings capability).
Choose a Period — Last 7 days, Last 30 days, Last 90 days, All time, or a custom range.
The report shows Total API Calls, Total Tokens and Estimated Cost, plus Cost by Feature, Cost by Model and a Daily Breakdown.

Costs are estimates based on published provider pricing at the time of each call; your provider’s invoice may differ slightly. This helps you understand where AI is being used most and control costs over time.

Other AI-assisted tools

AI also powers several tools documented on other pages:

Imported Drafts metadata suggestions and AI Vision Extract for scanned PDFs — see Bulk Document Import & Branded PDFs.
Framework control briefs, reference document classification and the per-document Compliance Audit — see Regulatory Framework Mapping.

Privacy and data handling

AI features are off by default. They must be explicitly enabled with the Enable AI features master switch (but see What the master switch covers for the three tools with their own controls).
Documents classified Confidential are never sent to any AI feature. This cannot be changed by any setting.
Documents with no confidentiality level set are also always kept away from AI. There is no setting to allow them — give the document a level first.
By default, documents classified Restricted are blocked too. Public and Internal Use documents are eligible.
You can change which levels are blocked under AI Content Privacy → Block External AI by Confidentiality (default: Confidential + Restricted). The checkboxes are Public, Internal Use, Confidential and Restricted; Confidential is always ticked and can’t be unticked. A document at a blocked level returns an error when any AI generation is attempted.
Allow AI audit on restricted documents (shown under AI Content Privacy when Regulatory Frameworks is enabled) controls whether the framework Compliance Audit may run on Restricted documents without a per-audit confirmation. It only takes effect if Restricted is not ticked in Block External AI by Confidentiality. At the default settings, Restricted is ticked, so neither this option nor the per-audit confirmation lets the audit run on a Restricted document.
The queue of background AI jobs briefly holds the AI’s results, which can include text drawn from your documents. Jobs are deleted about an hour after they finish.
Confidential documents are also left out when other documents are sent as background context for an AI request (for example, the catalogue used for related-document suggestions), not just when they are the document you’re working on.
When search intent rewriting is on, visitors’ search text is sent to your provider.
Your API keys are stored encrypted in the WordPress database.
Every AI call is logged locally with the feature, model, token counts and estimated cost.
Data sent to the AI provider is subject to the provider’s terms of service and privacy policy. Links to both providers’ terms and privacy policies are shown in AI API Configuration on the AI & Automation tab. The plugin also adds a description of its AI data flows to the WordPress Privacy Policy guide.

Limits at a glance

Limit	Value
AI calls per day, whole site	Daily API Limit, default 100 (10–10,000)
AI calls per day, per user	Per-User Daily API Limit, default 50 (0 = no cap)
When the daily counts reset	Midnight UTC
Short-term burst limit for Imported Drafts and Vision tools	120 requests per user per minute
How long the page waits for a background AI job	6 minutes
FAQs per document	1–20 (default 5)
Comprehension drafts per request	1–12 (default 6)
Document text read for comprehension questions	About the first 12,000 characters
Related-document suggestions shown	1–10 (default 5); suggestions remembered for 24 hours
Smart Search rewrites remembered	10 minutes

Every AI call also costs money with your provider. Check Analytics → AI Costs regularly, especially after bulk work.

Troubleshooting

“AI API key not configured.” No key is saved for the provider that matches the model in use. For example, you picked a Claude model but only entered an OpenAI key. Remember that per-feature models (such as Search Intent Model) can point to a different provider than the Default AI Model. Add the matching key under AI API Configuration, or choose a model from the provider you have a key for.

“Daily API limit of 100 calls exceeded. Resets at midnight.” or “Your personal daily AI limit of 50 calls has been reached.” The site-wide or per-user limit is used up. Wait until the counters reset at midnight UTC, or raise Daily API Limit / Per-User Daily API Limit under Usage & Limits. The Imported Drafts and Vision tools can also show “Too many AI requests. Please wait a minute and try again.” — wait a minute and retry.

“…blocked: confidentiality level “” is configured to prevent external AI access” (or any “blocked … confidentiality” message) The document’s confidentiality level is blocked from AI. Empty quotes mean the document has no level set, and documents without a level are always blocked. Set the document to Public or Internal Use if it’s appropriate to send it to your provider, or leave it blocked.

The button spins for a long time, then reports a failure or timeout Background AI jobs depend on WordPress’s built-in scheduler. If your host has switched it off without a server-side replacement, or the site gets very few visitors, jobs can sit in the queue until the page gives up after 6 minutes. Ask your host to make sure WordPress scheduled tasks run, then try again. A very long document with a slow model can also take a while; try a faster model for that feature.

“Invalid or unpublished document.” when generating a summary or related documents Summaries and related-document suggestions only work on published documents. Publish the document first, or use FAQ generation, which also works on drafts.

“The AI did not return any usable questions. Try again or reduce the requested count.” The AI’s reply couldn’t be turned into valid questions. Try again, ask for fewer drafts, or check the document has enough factual content to test.

Worked example

Ridgeview Aged Care is rolling out a revised Medication Management Policy and wants FAQs, a summary and comprehension questions without sending anything sensitive to an AI provider.

The administrator enters an Anthropic API key under AI API Configuration, keeps Default AI Model at Claude Sonnet 4.6, and switches on Enable AI features. She sets Daily API Limit to 200 for rollout week and leaves the per-user limit at 50.
Under AI Content Privacy, she leaves the defaults: Confidential and Restricted are blocked. The Medication Incident Register procedure is marked Confidential, so it can never be sent. The policy itself is marked Internal Use.
On the published policy, she clicks Generate Summary, then Generate FAQs. She confirms the consent dialog, unticks two suggested FAQs that duplicate existing ones, and clicks Create FAQs. Five draft FAQs appear under FAQs → AI Generated. She sorts by AI Confidence, fixes one 72% answer about after-hours dosing and publishes all five.
In Acknowledgement Content & Comprehension, she sets Drafts to 6 and clicks Generate with AI. She approves four questions, discards two that test trivia, and saves the document.
She clicks Suggest Related Documents. The AI proposes the Clinical Handover Procedure (high relevance) and the Incident Reporting Policy, with an audience warning that care staff can’t see the latter. She saves only the handover procedure.
Three months later, a minor version of the policy is approved. The automatic question review flags one question as Needs review because the dosing threshold changed, and suggests a new question about the new threshold. She sees the notice on the edit screen, clicks Remove on the outdated question, Approve on the new suggestion, and saves.
At month’s end, Analytics → AI Costs shows what the rollout cost, broken down by feature and model.

Version History

Version 1.0.0 Major October 4, 2026

Author: DHDW