Foundation

Set clear tasks and work with text, analysis and planning while retaining control over the result.

Choose any chapter. This level builds on earlier levels or equivalent knowledge.

What should be different when the task is finished?

In level 1, we learned to distinguish method, result and decision. We now move on to practical work. Our first question is not which tool to open, but what result we need.

“Use AI more” describes an approach, not a completed result. A goal might instead be that visitors understand the booking process, or that information becomes easier to compare. We can then assess whether the assistance actually works.

A short task description

In a fictional swimming association, adults need information before trying an activity for the first time. The facts are agreed: Sunday, 10:00–11:00, hall B, advance registration and free participation. The material includes no medical conditions or special exceptions.

A useful brief can identify audience, purpose and deliverable: a short information text for first-time visitors, clearly showing time, place and how to register. Here, brief means a consolidated description of the task.

Completion criteria might require all agreed facts, no additional conditions and an understandable next step. If the registration route is missing, list it as an open point rather than inventing an address.

This can be checked whether writing is manual, template-based or AI-assisted. The tool is therefore not the measure of success.

What the goal helps us exclude

The association does not simultaneously need a large marketing concept, an image series and a registration platform. Those would be other deliverables with different checks. Combining everything makes it harder to tell when the essential information text is ready.

A goal should be concrete without controlling every word unnecessarily. Understandable information is more useful than an exact sentence-length requirement that harms reading.

The same principle applies to customer messages, work instructions and meeting summaries. Audience and use determine what readers must be able to find and understand.

Keep this in mind

Start with the need, recipient and observable result. State what you will compare the result against before asking for a suggestion. That makes AI useful assistance without allowing tool activity to replace the task's real purpose.

What the tool needs to know

Context is information that makes a task understandable: audience, conditions, document extracts or existing decisions. Relevant context helps bound the task; large amounts of text are not automatically relevant.

Anthropic's guidance emphasises clear instructions and relevant information. We use that principle without assuming a particular product or guaranteed correctness from lengthy instructions.

Support for a work description

In a fictional gardening business, a supervisor wants a general preparation list for a customer day. The event is outdoors, an information area has been booked and participants must find the correct entrance.

Relevant context includes purpose, practical site conditions and agreed arrangements. An old customer list, personal employee notes and internal prices are unnecessary for organising the general list.

Instructions can therefore request use of these general facts, distinguishing confirmed preparations from pending decisions. If weather protection is unspecified, describe possible solutions as proposals rather than equipment already booked.

When information is important and sensitive

Sometimes needed information cannot simply be provided to a service. A legitimate task alone is insufficient. The organisation needs appropriate handling, permissions and approved tools.

EU data-protection guidance states that GDPR applies when personal data is processed in AI use. This chapter authorises no particular disclosure. Fictional and general examples let you understand the principle without private data.

A general description may explain the need better than an entire case history. This does not mean removing important conditions merely to shorten the instruction. The balance is the right facts, not the largest or smallest amount.

Keep this in mind

Context should improve relevance and make checking easier. Identify sources and flag missing or uncertain information. Unnecessary details should not accompany a task out of habit. More context does not mean unlimited data sharing.

The same message may need different wording

A technical colleague and a new customer may need different explanations. Audience adaptation concerns what readers need to understand and do, rather than replacing facts.

Isof's plain-language guidance emphasises finding, understanding and using information. This is a useful starting point for AI-assisted wording too.

Two versions of the same input

In a fictional bicycle workshop, a customer's inspection has moved to Thursday at 14:00, and the customer must confirm. The material gives no reason for the change.

Customer wording might be: “Your booked inspection has moved to Thursday at 14:00 Please confirm whether the time works for you.” Internal wording might be: “Inspection rescheduled to Thursday, 14:00 Customer confirmation remains pending.”

The first addresses the recipient directly; the second describes case status. Both retain the time and need for confirmation. Neither may invent illness as the reason or claim the customer already agreed.

Friendly wording must not make uncertain information certain. Technical wording must not hide facts the colleague needs.

What a role instruction does and does not do

“Write like an experienced customer communicator” can specify style. It does not create an employee with authority to change bookings or promise compensation.

Connect the role to a clear task: write understandably, use agreed information and flag unknowns. If specialist review is required, make clear that the tool's wording does not replace that assessment.

Audience adaptation matters in teaching, associations and public services too. It need not assume ignorance or make everything childish. Explain technical terms where needed and suit the situation.

Keep this in mind

Describe audience and use before tone. Compare language and meaning. Two versions may sound different yet remain equivalent when facts, conditions and uncertainty are preserved.

A result should be usable

An answer can contain correct information yet be difficult to work with. Specify its organisation: lists, short paragraphs and tables suit different needs.

Format should reveal meaning, not encourage guessing to fill every cell.

Comparing course materials

In a fictional craft course, material, quantity and cost information supports purchasing. The organiser requests columns for material, quantity, unit, unit price and total.

Fabric quantities are in metres; pens are counted as items. A quantity heading alone may conceal this difference. A clear specification includes units and the calculation used for totals.

A missing price should say, for example, “price missing”, not automatically become zero euro. Zero is known; an unknown price is an information gap. A complete-looking table may otherwise conceal the most important uncertainty.

Request only supplied information, retained units and visible completion needs. Recalculate totals independently when the values are available.

Headings and order

Format also concerns prose. If readers chiefly need time, place and next steps, put them early. Lengthy background can obscure essentials despite including everything.

Column names and headings should mean the same throughout. Price switching between total and unit price makes comparison misleading; more attractive colours cannot solve it.

Repeated formats need a simple definition of the fields. That becomes working information, not merely a visual sample.

Keep this in mind

Choose a format supporting the task and checks. Specify columns, units and unknown-value markers. Neat output is useful only when meanings and source material are clear.

When one assignment becomes too large

An assignment can combine collection, analysis, decisions and presentation. Mixing them makes errors difficult to locate. Separating them makes work and output easier to review.

The number of steps is not the goal. Each needs clear input and an output the next step can use.

A simple event report

1. Checked input
2. Analysis
3. Presentation

A fictional association wants to summarise an open day. It has confirmed attendance, cost information and general organiser notes.

First organise the material: which facts are agreed or checked, which estimated, and what is missing? Unreported costs must not become zero.

Next analyse available information. Calculate with known values and units. Read notes as observations, not automatically the views of all visitors.

Third present what the analysis actually supports. AI can clarify wording, but the report must not add confident conclusions about unresolved results.

Each step then has a check: sources and gaps first, calculations and reasoning next, meaning last.

Dependencies between steps

Confident presentation claims need checked evidence. You can prepare provisional headings early, but that differs from claiming all costs are verified.

If analysis reveals contradictory facts, return to the input. Stopping there is not failure: it prevents a gap becoming a finished mistake.

Other settings may involve materials, customer cases or teaching plans. Input, processing and presentation remain distinct. This also helps several contributors: each handover can state what is complete.

The presentation must depend on evidence that has already been checked. A provisional heading structure can be prepared early, but that does not justify stating that every cost has been verified. Returning to the sources when a contradiction appears is how a gap is prevented from becoming a finished error. The same separation can support material planning, customer cases and education. Where several people contribute, each handover should identify the completed work rather than leave the next person to infer its status.

Keep this in mind

Divide work by real dependencies. Later steps must not treat missing or unconfirmed information as available. Identifiable input and output help assess both AI assistance and human checks.

Describe what needs changing

A first answer often provides material for further work. “Make it better” may produce a different version without explaining the problem. Specific feedback points to an observable weakness and what the revision needs to achieve.

It also helps establish whether the change actually improved anything. Text that merely sounds different need not be more useful.

Three clear changes

In a fictional museum, information specifies a Sunday tour at 14:00, advance registration and an unresolved price. The constructed draft begins with lengthy history, says visitors can arrive whenever they wish and invents a price.

Feedback might request moving time and registration to the beginning, replacing flexible arrival with the agreed tour time, and removing the price while flagging confirmation.

These address different weaknesses: order, factual contradiction and unsupported information. They are more useful than merely asking for shorter or more professional text.

Compare the revision with those points. Time and registration should appear early, false flexibility should disappear and price should not appear known. Preserve other facts too, so correction does not create another mistake.

Change one thing without losing another

Rewriting can alter information you wanted to retain. State what remains unchanged: facts, audience and important conditions. A heading change need not regenerate the entire text.

A short change list can show differences and outstanding points. It does not prove correctness, but makes checking easier.

Repeated problems may require clearer input or assignment. More attempts with the same vague instruction are not always the answer.

Keep this in mind

Good feedback identifies weakness, desired change and what must remain. Compare the new answer with these and the original input. This supports AI-assisted revision without confusing variation with improvement.

A starting point, not final delivery

A draft makes form and content visible before completion. It may make starting easier, but still needs review and editing. Good first wording is not automatically publishable.

Separate agreed information from possible suggestions. Alternative headings may be appropriate; agreed dates must not vary as style choices.

Information for a craft course

Fictional example: Saturday, 09:00–12:00, workshop C. Participants bring an apron. No other material requirements are specified; contact details will be confirmed later.

The brief requests two short paragraphs for newcomers, clearly stating time, place and preparation. Mark contact details open and invent no safety instructions or equipment requirements.

A useful constructed draft can present agreed facts and separately note “contact route requires confirmation”. This working note must not accidentally be sent as finished participant information.

If the draft says safety glasses are provided, check it. Plausibility in a workshop is not evidence. This example does not decide a real course's safety needs; the responsible organisation does.

Review in two directions

First compare draft to input: additions, changes or excessive certainty? Then compare input to draft: missing essential facts?

The second is easy to overlook. Avoiding invention still permits omission of registration or location. Accuracy concerns what is stated and what needs stating.

After open points are clarified, create a final version. Retain the correct input and identify approved text, especially for several users.

Keep this in mind

A useful draft follows the brief and exposes uncertainty. Check additions and omissions, separate working notes from participant text and obtain appropriate factual confirmation before publication.

Easier reading without reduced accuracy

Clearer text can use simpler words and better order while retaining necessary details. Simplification misleads when conditions, exceptions or uncertainty disappear.

Isof's plain-language advice concerns finding, understanding and using information, not always using the fewest words.

A condition that must remain

Fictional venue booking: “The booking is provisional until the association confirms venue availability. Payment follows confirmation.”

A constructed simplification says: “Your booking is complete. Pay whenever you like.” Shorter, but it changes status and payment conditions.

A clearer version is: “Your booking is not yet confirmed. Once we confirm the venue is available, we will send payment information.” Everyday wording preserves confirmation before payment.

Do not add an exact payment date absent from the original. Clearer wording should not create new conditions.

What matters to preserve

Identify required actions, applicability and prerequisites. Explain technical terms rather than merely remove them.

“Check according to the current instruction” may need directions for finding it. “Check carefully” is not necessarily more usable.

Match the audience. An expert colleague may need a precise term; a visitor needs its explanation. Some sentences may become longer to clarify meaning.

Audience adaptation may sometimes make a sentence longer. A colleague with subject knowledge may need an exact technical term, while a visitor needs an explanation of that term. Removing terminology without explaining the underlying requirement can make the text less useful. The aim is not minimum length in isolation, but language that helps this reader understand the same conditions and take the right action.

Keep this in mind

Compare meaning before and after editing. Preserve essentials and explain difficulties. Shorter is better only if readers still understand the correct information.

A shorter account of defined material

A summary concentrates source material without silently adding conclusions or strengthening uncertain observations. Separate requested analysis or recommendations from the account.

AI can help structure and wording, but quality depends on comparison with the source.

Course-day information

Fictional note: “Twelve people attended. Four gave written feedback. Three of these wanted more practical time. Cost is provisional because the final venue invoice is missing.”

A constructed answer says: “All participants wanted more practice, and the course stayed within budget.” Both exceed evidence. Three of four respondents are not all twelve participants; provisional cost does not establish budget compliance.

A supported summary is: “Twelve attended. Four gave written feedback; three requested more practice. Cost remains provisional.” Counts, sample and qualification remain.

What is easily lost

May, provisional, according to the note and not yet can determine meaning. Removing them makes shorter text more confident than its source.

Keep contradictions visible. Conflicting times need clarification, not unexplained selection.

Requesting “the whole report” when only an extract is supplied does not provide missing pages. Output scope must match available material.

Checking claims

Every factual statement needs original support. Check omitted qualifications and central points too. Reading a plausible summary alone is insufficient.

Suggested next steps can appear separately as recommendations, avoiding confusion with original results.

Scope is part of accuracy. If a tool receives only a report extract, its output must not be labelled as a summary of the whole report. A heading does not give access to missing pages. Similarly, recommendations belong in a separately identified section: they may be useful, but readers must not mistake them for findings in the original. Check both the claims included and the central points or qualifications omitted.

Keep this in mind

Good summaries are shorter, not more inventive. Preserve sample, uncertainty and scope. Separate reporting and analysis so the recipient understands what the source actually says.

The same meaning in another language

Translations should suit readers and preserve meaning. Word-for-word replacement is not always enough. Adapt tone, terms and context without altering conditions.

AI translation is a draft. Compare important facts between versions and against the underlying information.

A small example with a large difference

Fictional course organiser: Swedish meaning is “The booking is provisional. Confirmation will be sent when the venue is ready.” A constructed English answer says: “Your booking is confirmed. We will send the venue details later.”

Natural wording changes provisional to confirmed. One correction is: “Your booking is provisional. Confirmation will be sent once the venue has been confirmed.” If Swedish klar means something else, clarify it before final translation.

A working glossary can pair preliminär bokning with provisional booking and bekräftad bokning with confirmed booking for this material. It is contextual, not universal.

What comparison covers

Check names, times, quantities, conditions and uncertainty. 15:00 may become 15:00 without changing the time. By and before can affect action requirements.

Back-translation alone does not prove correctness. Important translation needs sufficient language and subject expertise.

Calling the tool a “certified translator” guarantees neither competence nor formal authority.

Words that appear small can change an obligation. A deadline expressed as by a date is not necessarily identical to before that date. Time-format adaptation, such as 15:00 becoming 15:00, must retain the actual time. If the original phrase about the venue being ready remains ambiguous, record the ambiguity and obtain clarification. Do not settle it simply because one translation sounds fluent. The contextual glossary supports consistency without claiming that every word always has one English equivalent.

Keep this in mind

Use consistent terminology and compare meaning, not word count. Flag ambiguous originals rather than guess. Natural translation should give equivalent factual conditions.

Many suggestions are not the same as a good choice

AI can offer alternatives when work stalls. A long list does not establish suitability. Assessment needs clear criteria and appropriate conditions.

Define the desired outcome before seeing suggestions, so attractive wording does not control the choice.

Five open-day ideas

A fictional cultural centre wants new visitors to find activities. Budget is small, two staff are available and the event is indoors.

Options are a welcome sign, venue map, staffed information point, new mobile app and hourly tours.

Compare orientation, preparation and staffing. A sign is simple but may not show the entire venue. A map gives overview but needs correct layout. A desk answers questions but uses staff. An app may require work beyond this small task. Tours commit staff at fixed times.

Signs and maps may take priority, with the desk assessed against staffing. The app is not first just for modernity. This is a reasoned fictional choice, not a universal rule.

What else needs establishing?

An unfinished floor plan means the map cannot be considered ready. Unknown visitor numbers must not become invented staffing figures.

Unknowns may justify investigation, not confident good/bad labels. Make assumptions visible.

Changed goals may require different criteria. Accessible information for a particular group deserves explicit assessment, not concealment behind general attractiveness.

Unknown visitor numbers should not be invented to make staffing comparisons look complete. An unknown fact can justify investigating an option without proving it good or bad. The map is not finished if the venue plan is unsettled. If the goal later changes to accessible information for a particular visitor group, assess that requirement explicitly. General attractiveness cannot stand in for the information that group needs.

Keep this in mind

Let AI broaden ideas, then compare against shared criteria and actual boundaries. A small chosen solution with documented outstanding work may be useful. Idea count does not measure value.

A plan describes relationships, not just dates

A neat task list may still be impossible. Identify each step's input and when it can exist.

Duration is often estimated, particularly for new work. Label estimates rather than use precise numbers to imply certainty.

Two dependencies to correct

A fictional plan publishes an invitation Monday, confirms venue Tuesday, orders signs Wednesday and finalises entrance information Thursday.

The invitation may need a confirmed place. Monday's confident announcement is out of sequence. Preparing provisional text is different from approved sending.

Sign content depends on entrance information. Format and quantity may be planned early, but content needs confirmation before printing.

Separate preparation from confirmed delivery: confirm venue before stating it as settled, and entrance information before ordering signs.

What can happen in parallel?

Headings and general layout may develop during venue checks, leaving location open. This allows parallel work without treating missing decisions as complete.

AI can suggest order and dependencies, which need actual-process comparison. It does not know external confirmation times without information.

Optimistic-only plans are vulnerable to waiting and corrections. Necessary review, completion and handover are not waste.

Parallel preparation is possible without parallel decisions. Heading suggestions and a general layout can be developed while the venue is checked, provided the fixed location remains open. Review, completion and handover also need time. A plan containing only optimistic durations is vulnerable to waiting and correction. These necessary steps are not waste simply because they do not immediately produce the participant-facing message.

Keep this in mind

Distinguish possible preparation from delivery requiring confirmed input. Plan dependencies and disclose time uncertainty. Calling every step parallel does not supply missing facts.

What was said, decided and assigned?

Notes can mix ideas, questions and decisions. Careless summaries turn proposals into commitments nobody made.

AI may organise them, but meaning needs checking. Recording and personal-data handling follow organisational decisions and rules. Here we use constructed notes.

Three types of content

Fictional note: “Someone proposed another information sign. The group decided to retain the entrance. Venue coordinator checks the old map. No date agreed.”

Separate discussion proposal, entrance decision and assigned map check. Date remains undecided, not next Friday to complete the list.

“Coordinator orders a new sign by Friday” changes action and date without support.

Why headings matter

Discussion is raised; decision is adopted through the meeting's process; action is concrete next work with responsibility and dates where assigned.

Important open questions are not already resolved. Flag unclear points for appropriate confirmation.

People absent from the meeting need applicable status, not plausible conversation.

Check at handover

Compare decisions/actions against notes. Were action, owner and date actually agreed? Check omitted open questions.

Identify current summary among versions. Old lists should not remain active after correction.

The headings carry status, not just presentation. Discussion means a topic was raised. Decision means a position was actually adopted through the meeting's decision process. Action means concrete further work with assigned responsibility and a date where one has been agreed. An important unresolved question is still unresolved. This distinction matters especially to someone absent from the meeting, who must act on applicable information rather than plausible recollections. At handover, compare every stated action, owner and date with the notes, and check whether an important open question disappeared.

Keep this in mind

Organisation should clarify status, not invent decisions or responsibility. A short list with visible gaps beats apparent completeness built on invented information.

A friendly reply must also follow the facts

AI can support quick accommodating wording without adding promises.

Review needs verified case facts and rules governing what you may promise. A language model should not determine customer rights.

A fictional rescheduling

A bicycle workshop can offer Thursday at 14:00, not yet accepted. Its fictional procedure prohibits costs/compensation promises without responsible approval.

“Your appointment is Thursday at 14:00, free as compensation” converts option to confirmation and invents authority.

Supported wording: “We need to reschedule your inspection and can offer Thursday at 14:00 Please let us know whether that works for you.”

This invented procedure assesses no real statutory customer rights.

What information is needed?

Drafts often need only invented names/general case facts. Unnecessary personal details should not be shared by habit. Necessary real details require approved handling.

Separate internal notes from customer text. “Compensation needs deciding” is not an agreed condition and must return to the appropriate owner.

Check more than tone

Compare times, status, prices and conditions. Confirmed, can offer and will convey different meanings. Gentle words may carry substantial promises.

Use the approved version, not an old draft conveniently near the top.

A working note about compensation needing a decision should not be sent as if compensation has already been agreed. Return that question to the responsible person first. Wording such as confirmed, can offer and will describes different states, even when all three sentences sound polite. Compare booking status, prices and conditions as well as times. Once the message is approved, identify the version to send; an earlier draft must not be used merely because it is conveniently visible in the history.

Keep this in mind

Useful customer replies are understandable, grounded and within authority. Friendly wording cannot replace factual checks.

What do rows and columns say?

Tables condense information without making meaning obvious. Establish measures, units and blanks.

AI may help questions and summaries; interpretation still needs table comparison.

Three records

Fictional course days:

A: 12 booked, 10 attending, materials € 300.

B: 8 booked, attendance unknown, materials € 240.

C: 15 booked, 13 attending, cost missing.

C has 15 bookings, not 13. Known attendance A+C is 23; total attendance is unknown because B is absent.

Known cost A+B is € 540, not the full period or cost/person. Questions determine relevant rows/columns.

Unknown is not zero

Zero attendance for B invents a fact. Known-cost subtotal is possible if scope is explicit.

Google identifies missing values, wrong units and unclear definitions as recurring data issues. Tables do not eliminate these.

“Higher-cost days always have more participants” exceeds this incomplete material. Association differs from causation; three rows cannot justify every generalisation.

Make checks possible

State row, column, unit and gaps so others can compare.

Do not interchange booked and attending to simplify answers. Discovering an unanswerable question may be valuable.

The question determines which rows and columns can support an answer. A subtotal of known costs can be useful if its scope is stated; it is not the whole period's cost or a cost per participant. Similarly, booked and attending must not be interchanged simply to answer a question more easily. A constructed claim that higher-cost days always have more participants goes beyond these incomplete records. Association and causation differ, and three rows do not support every generalisation. Showing the supporting row, column, unit and missing values makes the interpretation reviewable.

Keep this in mind

Begin with meanings and units. Flag gaps and distinguish subtotal from complete result. Explanations should trace to values used.

A polished answer can contain errors

AI can help describe a calculation, suggest which items are needed or explain a formula. But clear wording does not prove that the result is right. Check the calculation using a separate method, such as a calculator, spreadsheet or your own working. Also check whether the method fits the question.

First establish what the numbers mean. Is the amount for one person, one hour or the whole group? Do the prices include the same things? Writing out units often reveals arithmetic errors, while an incorrect assumption can survive even correct multiplication.

A budget you can follow

In a fictional example, an association organises a workshop for 18 people. Materials cost € 45 per person, and the venue costs € 600 for the session. The budget includes only these items. Materials cost 18 people × € 45/person = € 810. Adding the venue gives 810 + 600 = € 1,410. Divided among 18, this is approximately € 78.33 per person.

If an AI draft states that venue hire costs € 600 per person, it has changed the conditions. The arithmetic may be correct but answer the wrong question. Clarify that venue cost covers the whole session. Travel, refreshments and other possible costs are excluded; the total must not be presented as covering everything that may be needed.

Plausibility before precision

A rough estimate helps: materials are close to € 800 and venue hire is 600. The total should be near 1,400, not 14,000 or 140. This does not replace calculation, but exposes large errors. Round when needed and distinguish exact amounts from rounded estimates.

If attendance changes, recalculate from the input. The previous total is not a valid basis for every new situation.

Retain the trace back

Useful results show input values, units, method and exclusions. Colleagues can understand and check without trusting the answer's confidence. Decisions with financial consequences retain the organisation's normal checks and responsibility.

Decide what the comparison should help with

AI can quickly produce a comparison table. The difficulty is choosing relevant properties and treating alternatives equally. Assessing one by price and another by positive reviews produces a misleading comparison.

Begin with the decision: what must work, and for whom? Separate mandatory requirements from preferences before an option becomes attractive. Low price cannot compensate for missing accessibility required by the event.

Ask every option the same questions

1. Same criteria
2. Facts / unknown
3. Reasoned choice

A fictional cultural association needs space for 25 people. A costs € 900, holds 30 and has confirmed step-free access. B costs € 700 and holds 25, but access is unknown. C costs € 1,100, holds 40 and has confirmed step-free access. Prices cover the same period; equipment and cancellation terms remain unchecked.

Ask AI to organise the same columns: price, capacity, access and outstanding questions. A and C meet known capacity/access requirements. B needs more information. That does not mean its entrance is poor, only that the fact is missing.

Make facts and values visible

A being cheaper than C is factual within the example. A being best is a judgement depending on other requirements. More space may favour C. If both suffice and other conditions are equal, price favours A. State recommendation conditions.

Scoring and weighting may help, but do not automatically make conclusions objective. Unknown properties should not receive invented average scores to complete the table. Identify what needs investigation.

A comparison should be revisable

Keep date and source information, especially when prices or availability change. A brief account of what new fact could alter the conclusion is useful. The comparison then supports decisions with visible limits rather than an apparently certain ranking.

An image should explain something

An illustration brief states what the image should help readers understand. Begin with the learning goal and the adjoining passage. Identify which relationship, distinction or sequence becomes clearer visually. If the picture merely fills space, it need not exist.

Daynamic Learning requires original illustrations produced for the course. Internet images, stock images and copied diagrams are excluded. Original production still needs documentation; generation alone establishes neither every right nor absence of resemblance to existing work.

From explanation to subject

A fictional proposal uses a small illustration showing known versus unknown in a comparison. Two neutral venue cards show the same three properties. One shows confirmed price and access; the other's access is visibly unknown. Missing facts require investigation rather than guesses.

The brief calls for simplicity, readability at small size, limited detail and meaning not conveyed by colour alone. Avoid logos, real people and specific artists' styles. Labels may need separate typesetting and checking because generated text is not guaranteed correct.

Specify use

State placement, approximate size and removable detail if the image becomes crowded. Check print: thin lines and weak contrast can disappear. It supports text and must not carry essential information absent from the chapter.

Alternative text describes information: “Two comparison cards show confirmed access for one venue and access still unknown for the other.” Decorative colours and shapes need not be listed.

Document and review

A planned name is KAP_038_kant_och_okant_01. Keep brief, production method, version, alternative text and status in Illustrationer. This is a proposed filename; no image has been produced in this work order. Review subject, text and pedagogical clarity before use.

Reuse the method

A template retains what usually matters: task structure, desired format and checks. It should not accidentally carry over another customer's details, an old time or a previous decision.

Separate fixed instructions from variables. Fixed instructions may require clarity, supplied facts only and marked unknowns. Variables change with each task: audience, purpose, facts, length and channel.

A fictional example

An organisation frequently announcing changed times can use:

Purpose: [What the recipient needs to understand or do.]

Audience: [Who the message addresses.]

Confirmed facts: [Time, place, change and other relevant information.]

Unknown or pending: [Undecided points.]

Format and tone: [For example, a brief friendly, factual email.]

Boundaries: Add no promises, prices or decisions absent from the input.

Check: Identify facts requiring confirmation before use.

Brackets show fields to complete, not text to leave in a finished message. Input must also comply with rules for the chosen tool.

When the template does not fit

Customer-information templates are not automatically suitable for sensitive staff cases or instructions where errors may cause harm. Adapt task and responsibility. “Thank you for understanding” may need changing when recipients have not received a clear explanation.

Keep the template proportionate. Lengthy instructions for each short message become hard to maintain. Include useful elements and update recurring weaknesses.

A template for customer information does not automatically suit a sensitive staff matter or an instruction where mistakes can cause harm. Adapt the task and its responsibility. A stock phrase such as thank you for understanding may be inappropriate before the recipient receives a clear explanation. Keep templates small enough to maintain: a long instruction for every brief message can become burdensome. Update them when recurring weaknesses are found rather than allow old details to become permanent defaults.

Finish with human checking

Check completed variables, absence of old details and agreement with current input. Save a version and short change note. Templates support consistency, not automatic approval of each output.

See how the parts connect

A worked fictional example combines the level's methods. No separate project is required to continue. The optional conclusion follows.

A library introduces digital services on Wednesday, 15:00–16:00, in group room B, for up to 12 people, free with advance registration. It addresses adults getting started. The registration route is unconfirmed. This is constructed teaching, not a real event.

Brief and first draft

The task is a brief understandable message. Criteria require correct time, place, audience and registration, without unconfirmed additions. Scope is writing, not publishing, booking or deciding.

An illustrative first draft says: “Drop into the library on Wednesday! We help everyone with digital services all afternoon. No registration needed.” Welcoming wording changes important facts: vague time, expanded audience and removed registration. This is an invented faulty draft, not actual tool output.

Check and correct specifically

Compare every input line. Request 15:00–16:00, group room B, adults getting started, required registration and no promise of every digital service. Flag the missing route internally.

A corrected suggestion: “Would you like to get started with digital services? The library holds an introduction for adults on Wednesday, 15:00–16:00, in group room B. It is free, with space for up to 12 participants. Advance registration is required.” The route remains open, so the proposal is reviewable but not publishable.

Complete under appropriate responsibility

Add and check the route once a responsible person confirms it. Preserve input, correction and final version to explain changes. The owner then decides use.

The workflow is bounded task, relevant information, draft, checks and corrections, confirmation of gaps. AI has a purpose while the organisation retains facts and usage control.

The optional level conclusion, explanatory feedback and shared sources follow. The next level begins with chapter 041 Map a workflow.

This is the level's optional conclusion. Choose useful activities or continue reading directly. There is no pass mark, submission requirement or barrier to the next level. All situations below are fictional.

European edition 1.0 | 2026-10-13

Optional further reading. EU sources supplement the original bibliography. National sources are supplementary examples, not rules for all Europe. Original sources were reviewed on 2026-10-12; EU additions on 2026-10-13. Publication titles and addresses are retained.