Setting up your practice2. Customise your cards

Setting up your practice

Customise your cards

Stage 2: use AI to draft practice-specific wording from your questionnaire, or choose to keep your existing wording.

  • Practice admins
  • Clinical reviewers

This stage prepares your cards before clinical review, so your reviewer checks the wording your team will actually use. You have two routes:

  • Customise with AI — AI rewrites your cards using your questionnaire answers. Every rewritten card then needs clinical review.
  • Use existing wording — you skip AI and your reviewer checks and edits the current cards by hand.

This stage only appears if the Signposting Toolkit team has switched AI customisation on for your practice.

Pick up where you left off

When you return, the setup panel shows your saved progress and outstanding reviews. Select Continue unfinished generation to open the saved run. Another practice admin can continue the same run.

Open Cards still to generate to check what will be continued. Saved cards are excluded, and later edits or review decisions are checked again before each card changes. If two admins confirm the same cards together, they join the same saved run.

While a run is unfinished, you can’t start a different selection. The AI customisation page shows Open saved progress: continue the saved run, or keep the current wording for its remaining cards, then choose new cards.

Cards already saved are kept. If the AI service is busy, the page shows a waiting message and retries automatically while it stays open. You can select Pause after this card, or leave the page and continue later. The current card may finish; the remaining cards wait for you to return.

If AI customisation is switched off, the return panel offers Check saved generation. Open it and select Keep current wording and review to finish the saved run.

If a card was edited or reviewed after selection, it is held back with an explanation. So is a card removed from the library, or one with no instructions to customise, because retrying wouldn’t help. Changed practice answers also stop unfinished generation. Check the affected cards before making a new selection.

Select Keep current wording and review if you want the remaining cards to keep their current wording. The confirmation explains what will be kept. Every outstanding card still needs clinical review.

Returning to an older generation run

Older runs did not save their full progress. Check remaining customisation lists only untouched shared cards that can safely be offered again. It excludes saved drafts, local edits and clinical review activity. Read the list before choosing to generate; the app cannot reconstruct the original selection.

Once generation is finished, the return panel shows how many cards are approved and what’s left to review. Why didn’t AI rewrite every card? explains that, unless you choose otherwise, AI leaves out cards someone has already reviewed, so fewer cards may have AI drafts than are in your library. Older runs didn’t record which cards they left out, so the app can’t list them.

Who does it

A practice admin runs AI customisation, ideally with the clinical reviewer on hand to judge the sample cards. Choosing Use existing wording also needs a practice admin.

Customise with AI

Finish stage 1 first. The AI card on the setup page can’t be opened until your questionnaire is finished and at least one appointment type is set.

On the setup page, choose Start AI customisation. The page, Make your cards fit your practice, has three preparation steps: Practice details, Sample preview and Generate cards. Confirming your selection opens your saved generation progress.

Check your practice details

  1. Check the appointment names, clinicians, descriptions and booking routes shown. They come from your questionnaire.
  2. If anything is wrong, choose Edit practice details or booking routes, correct it and come back.
  3. Tick These booking routes reflect our practice.
  4. Choose Continue to sample preview.

Ticking the box confirms your arrangements. It doesn’t change any saved card or approve any clinical content.

Preview sample cards (optional)

  1. Choose Preview up to five sample cards. Examples appear one at a time, with a progress bar at the bottom of the screen. Keep the page open while they are prepared.
  2. Check the overall approach: local appointment names, who patients should see and how your services are described. Each sample shows the important notice and any support panels, plus a For your reviewer to check list where automatic checks found something.
  3. If the same appointment or routing problem appears in several samples, choose Edit your practice answers, fix the answer, then generate a new preview.
  4. When the approach looks right, choose Looks broadly right — continue. Or choose Skip preview and continue.

A skipped or incomplete preview is labelled Skipped · optional or Incomplete · optional, rather than ticked.

Giving feedback on sample cards (pilot practices)

Some practices are in a pilot that lets you correct your practice answers straight from the samples.

  1. Choose Something needs adjusting, or Give feedback on this example under a sample.
  2. Describe the practice arrangement the samples got wrong. Use practice information only — no patient details.
  3. The AI may ask a short question. It then proposes a change to your questionnaire answers, showing the current answer next to the proposed one and which cards it applies to.
  4. Read the whole proposed answer. Adjust suggestion lets you edit it; edited suggestions are checked again (Check edited suggestion) before you can save.
  5. Choose a symptom under Example to include when previewing again if the current samples don’t show your change.
  6. Choose Save and preview again. This saves the new answer and generates fresh samples.

Use Previous and Updated to compare the samples. Fresh AI wording can vary beyond the change you asked for. Support panels are carried forward rather than rewritten.

Saving feedback doesn’t rewrite existing cards or change their approval. The new answers are used when you generate cards later.

Recent practice-answer changes lists who confirmed each change, with the previous and saved values. Undo this change is available only while no later questionnaire changes have been made.

If another admin changes the answers while you are working, you are asked to reload them. Check your suggestion again afterwards. Unsaved feedback and sample comparisons aren’t kept; confirmed answers and their history are.

If the answers save but the new preview fails, the message says so and you can try the preview again. The saved answer isn’t undone.

Generate your cards

  1. Under Which cards would you like to customise?, choose a scope:
    • Customise all symptoms — starts with your full library. This is the default, including for practices that already have local wording.
    • Select symptoms manually — search for and tick a smaller group.
  2. Check Cards excluded by default. Approved cards, cards with changes requested and other cards with human reviewer activity or notes are left out unless you tick to include them. The selection summary shows how many cards are Included for rewriting and how many are Excluded by review choices.
  3. Choose Review selection. This opens a confirmation. Nothing is generated yet.
  4. Check the lists in the confirmation. To change anything, choose Back to selection.
  5. Choose Replace wording on N cards to start.
  6. Keep the tab open. Cards are saved one at a time, and the progress panel shows the card being worked on. Leaving the page stops the remaining cards.

Each rewritten card is marked Pending clinical review. An earlier approval doesn’t carry over to the new wording.

How review exclusions work

The exclusions apply whichever scope you choose. Picking an approved card in a manual selection doesn’t override its protection — you still need to tick the matching inclusion.

  • Notes or earlier review actions don’t necessarily mean a card has been fully reviewed or approved.
  • Notes left by AI during earlier setup don’t count as human review, and don’t exclude a card on their own.
  • A card can have different review statuses across its age groups. Every relevant inclusion needs ticking before that card is included.
  • No human review recorded in the lists means no human review decision, activity or note is recorded for that card. It isn’t a clinical approval.
  • Including a card with a change request keeps the reviewer’s note visible and returns the card to Pending clinical review. AI uses your practice answers to rewrite the card; it doesn’t use the note as a correction instruction or mark the request resolved. Check the note against the new draft.
  • Earlier decisions, reviewers, dates and notes are kept under Reviews preserved before AI customisation in the clinical review preview, including after a later approval.
How the counts in the confirmation work

The selection page and the confirmation show the same summary.

  • Included for rewriting is the total number of cards in this run. Any count of included cards with review records or notes is part of that total, not an extra group.
  • Excluded by review choices is a separate group. Those cards keep their wording and review status.
  • Cards outside a manual selection are shown separately and are also left unchanged.
  • The counts beside each inclusion tick box can overlap. The final summary counts each card once, and its expandable lists show each card’s status or combination of statuses.

If nothing is included, a message explains that the review exclusions protect the whole selection. Every scope has the final confirmation, even a single manually selected card. Closing the dialog or pressing Esc leaves your cards unchanged.

If someone edits a card during the run, or you reload review choices

If someone changes a card’s wording or review after you made your selection, or while AI is working on it, that card may be skipped so their changes are kept. Check the results, then decide whether to include it again.

Reload review choices checks the latest review records. It clears your inclusion ticks but keeps your chosen scope and any manually picked cards. It doesn’t include reviewed cards on your behalf.

Cards left out keep their wording and review status. Disabled cards stay disabled.

A run also stops if your practice answers change while it is running, and it tells you which cards remain so you can continue deliberately.

What gets rewritten, and what is kept

For each included card, AI rewrites the brief instruction and the main instructions using your saved answers. It may adapt appointment names in an important notice, but must keep its emergency meaning.

The saved local copy keeps the card’s existing notices, related links, support panels and age-group versions, including ones left intentionally blank. Support panels and age-group text aren’t rewritten, so your reviewer should check them for local names too.

Where no suitable appointment is set up, the wording keeps the original urgency and uses general escalation wording. A same-day route shouldn’t become routine or 48-hour care.

Automatic checks flag likely problems for your reviewer — for example missing same-day or 999 wording, changed patient links, placeholders, or appointment colours your practice doesn’t use. They are advisory, not a full check of clinical meaning. The flags appear after generation and are saved in the card’s review note.

Check the results

The saved progress page shows Cards saved, Still to generate and Need attention.

  1. Read Cards needing attention or kept unchanged for any cards that were held back.
  2. Select Retry failed cards for unsuccessful requests. Cards changed or reviewed since selection remain protected.
  3. Choose Review your cards once generation finishes. Advisory checks are saved in each card’s review note for your reviewer.

If nothing is left to generate but some cards were held back, select Keep current wording and review to finish the run. The held-back cards keep their current wording for clinical review.

Cards excluded before the run are not failures. You can return to this progress page after closing the app.

Smart re-run (system administrators)

System administrators see a third scope, Smart re-run. It compares each card’s current brief and main wording with what AI last produced, leaves out cards it identifies as locally edited, then applies the same review exclusions. Edits only to an important notice aren’t detected this way, so check the proposed scope carefully. The saved run keeps this protection when you continue or retry it.

Use existing wording

If you would rather review your current cards than generate new ones:

  1. On the setup page, choose Use existing wording on the Customise your cards with AI card.
  2. Read the dialog and choose Confirm existing wording.

The card then shows Using existing wording · AI skipped. Your questionnaire, cards and review decisions stay as they are. Every card still needs clinical review.

You can undo the choice with Change this choice and then Return to AI customisation. If you run AI after review has started, it may replace wording your reviewer has already checked, unless you leave those cards excluded.

The same choice is offered in a notice at the top of clinical review, alongside a link to customise with AI, if you open review before this stage is done. Customise manually instead on the AI page’s first step takes you straight to clinical review, but doesn’t record the choice.

What done looks like

The stage is complete when either of these is true:

  • AI customisation has saved new wording on at least one card
  • your practice has chosen Use existing wording

The card shows AI drafts generated or Using existing wording · AI skipped. Opening clinical review, approving cards or requesting changes doesn’t complete this stage on its own.

Common questions

Do AI drafts go live straight away? Pending is a review status, not a visibility setting. Pending cards that are switched on stay visible to staff, with a warning, until they are approved. See Review your cards.

Should we preview samples before every run? It is optional. It is most useful the first time, or after you change your practice answers.

Can we re-run customisation later? Yes. See After go-live.

Many cards were skipped. First separate cards excluded before the run from cards that failed during it. The results give a reason for each unsuccessful card, such as no instructions to customise, changed wording or reviews, or a failed request.

AI added a routing step that wasn’t in the original. Don’t assume it is right because AI produced it. Your reviewer checks it against the original card and your agreed rules, and edits it or requests changes before approving.

Related: Using AI safely · AI drafts and smart visuals

Next: Review your cards.