Capability / AI workflows
AI-assisted workflows for tourism teams
A revised visitor programme arrives from several partners. Your team needs to identify what changed, prepare useful summaries and update different language versions. AI can help prepare that work when it receives relevant sources and a clearly defined task.
Tamaga designs AI-assisted workflows around the material you maintain and the work you need to complete. The aim is to make preparation useful and review manageable, while keeping factual decisions and publication under your control.
Start with work that repeatedly needs preparation
An operator, destination, cultural route or accommodation may have different material to manage. The useful starting point is a recurring task with identifiable inputs and a result someone can check.
Some changes need ordinary publishing rules rather than AI. Reusing a maintained date across connected pages, for example, should not require a model to invent or interpret it.
| Work to prepare | Useful AI assistance | What needs checking |
|---|---|---|
| Review existing content | Compare passages and identify possible repetition, conflicting explanations or unanswered questions. | Whether the pages actually serve different reader needs. |
| Prepare language versions | Draft a translation using the source, terminology and editorial guidance. | Meaning, names, conditions and natural expression. |
| Organize partner material | Extract proposed details and identify statements that need supporting evidence. | Sources, dates and who can confirm the information. |
| Find relevant information | Use semantic search to retrieve material by meaning, including differently worded questions. | Whether the retrieved passage answers the question and applies to the situation. |
Keep the condition when preparing another version
Consider a fictional cultural association whose programme says:
During October, studio visits can be arranged on Saturday mornings. Contact the association to discuss a date.
A proposed newsletter introduction says:
Visit the studio every Saturday morning in October.
The shorter version has changed the offer. Readers could reasonably understand it as a regular visiting schedule.
A useful workflow presents the source beside the proposal so the editor can retain the arrangement condition:
Interested in an October studio visit? Contact the association to arrange a Saturday morning.
The newsletter introduces the opportunity and leads to the maintained programme. It does not need to repeat the whole page, but it must preserve what matters to someone planning a visit.
Selected passage and retained condition: the excerpt stays connected to the material that qualifies it.
Conceptual illustration of a passage kept in context.
Build the workflow around an inspectable result
We begin by defining the task, the material it may use and the result the editor needs.
The implementation can fit an existing editor and revision history. A single editor should be able to understand what changed and decide what to save without adopting an elaborate approval process.
-
Step 1: Choose the inputs
Identify the relevant records, passages, terminology and dates. Decide what information may be sent to the selected AI service.
-
Step 2: Define the proposed output
A comparison, translation draft or suggested summary needs a clear scope. Missing information should remain identifiable.
-
Step 3: Make review practical
Show the proposal with its source and editing location. Let the editor keep, amend or reject it without reconstructing the whole task.
-
Step 4: Check representative examples
Examine useful results, consequential mistakes and failure behaviour before extending the workflow.
Check meaning as well as fluency
In Tamaga’s October 2026 editor tests, a tone rewrite changed “can also help” to “also helps.” That small alteration strengthened a possibility into an assertion.
A readable result can still change a condition, attribution or degree of certainty. Review therefore needs to match the consequence: a suggested tag and a statement about physical access require different attention.
When the supporting material itself is uncertain, Source Confidence helps establish what it supports. Finding a source passage is useful preparation; it does not settle every factual question.
The hotel example in What a chatbot should never answer from memory develops this distinction for consequential visitor answers.
Source and proposed version: kept available for comparison. A fluent rewrite still needs a check of conditions and certainty.
Conceptual illustration of an inspectable proposal.
Connect AI assistance to the publishing system
Where Drupal holds the content, Drupal AI provides a foundation for connecting model services and integrating assistance into editorial work. Its ecosystem includes tools for translation, content suggestions and semantic search. The appropriate combination depends on the task and existing site. Drupal AI project documentation
Our Drupal capability explains the wider implementation approach. Multilingual Content addresses how language versions retain their meaning and remain maintainable after the first translation.
One source, two proposed treatments: a fuller version and a shorter one remain distinct from the maintained material.
Conceptual illustration of alternative editorial treatments.
Bring one recurring task and representative material
Show us what arrives, what you currently do with it and what a useful result would look like. We can examine where AI assistance fits, what should remain a straightforward system function and which checks the editor needs.