AI Marketing and Productivity Guide

AI assistants have become an everyday tool for many small businesses, freelancers and marketing teams. Used well, they shorten the blank-page stage, speed up routine writing and help you think through problems. Used carelessly, they produce generic copy, confident mistakes and privacy problems.

This guide is about the careful version. It covers where AI helps, where it does not, how to write prompts that produce useful first drafts, how to review what comes back, how to handle data responsibly, and how to measure whether you are actually saving time.

One principle runs through everything below: AI output is a draft, not a finished product. It is not guaranteed to be original, accurate or undetectable, and you remain responsible for anything you publish or send. Every workflow in this guide includes a human review step for that reason.

What AI assistants are good at (and what they are not)

Before building workflows, it helps to have a realistic picture of the technology. Large language models generate text by predicting likely continuations based on patterns in their training data and whatever you give them in the prompt. That makes them strong at some jobs and unreliable at others.

Tasks where AI tends to help

• Getting past the blank page. Outlines, first drafts, alternative headlines and subject lines.
• Transforming text you already have. Summarising, shortening, changing the format (a blog post into social posts), changing the reading level, or restructuring notes into a document.
• Brainstorming options. Angles for a campaign, objections a customer might raise, questions for a survey, names for a workshop.
• Structuring messy input. Turning a meeting transcript into action items, or grouping open-ended survey answers into themes for you to check.
• Explaining and teaching. Explaining a formula, a marketing concept or a spreadsheet function, which you then verify.
• Drafting routine communications. Follow-up emails, polite declines, appointment reminders and support replies that a person then checks and personalises.

Tasks where AI is weak or risky

• Facts, figures and citations. Models can invent statistics, studies, quotes, URLs and product details that sound plausible. This is usually called hallucination.
• Anything recent. A model's built-in knowledge stops at a training cut-off. Unless the tool is connected to live search (and even then), treat information about recent events, prices, laws and platform features as unverified.
• Your specific business. The model does not know your prices, policies, customers or history unless you tell it, and it may fill gaps with guesses.
• Judgement calls with consequences. Legal, financial, medical, employment and safety decisions need qualified people. AI can help you prepare questions, not replace advice.
• Original insight. AI reflects common patterns. If your content needs a genuine point of view, first-hand experience or new data, that has to come from you.
• Arithmetic and data analysis without checking. Some tools calculate reliably when they run code; others approximate. Always check numbers.

A quick decision framework

Use this four-question test before handing a task to an AI assistant:

1. Is the input something I can safely share? If it contains confidential or personal data and you have not checked the tool's terms and settings, stop (see the privacy section).
2. Can I verify the output quickly? A summary of a document you have in front of you is easy to check. A list of "industry statistics" is not.
3. What is the cost of an error? A rough brainstorm has a low cost of error. A price quote, a legal statement or a health claim has a high one.
4. Will editing the draft take less time than writing it myself? For short, personal messages, the answer is often no.

If the answer to 1 is no, do not use the tool for that input. If 2 is no and 3 is high, do the task yourself or with a qualified person. Otherwise, AI is a reasonable place to start.

Realistic AI use cases for marketing and operations

This section walks through the jobs where small teams most often get value, with a practical note on each.

Research and planning

AI can speed up the early, exploratory phase of research: listing questions to investigate, suggesting ways to segment an audience, or summarising a long document you provide. It should not be treated as the source of facts.

A useful pattern is "AI for the questions, primary sources for the answers." Ask the assistant what you should find out about a market, a competitor category or a customer segment. Then answer those questions using sources you can cite: your own sales data, customer interviews, official statistics offices, regulator pages, company websites and reputable publications.

Example: a local bookkeeping firm planning content for sole traders might ask for "20 questions sole traders commonly have in their first year of business." The list is a starting point. The firm then checks which questions actually appear in its inbox and in search tools, and answers them with accurate, jurisdiction-specific information.

Drafting content

For blog posts, landing pages and newsletters, AI is most useful for structure and first drafts. The AI blog outline generator can produce a starting outline from a topic and audience, which you then reorder, cut and expand with your own experience.

Good drafting practice:

• Give the model your own notes, examples and opinions as input, rather than asking it to write from nothing.
• Draft section by section rather than asking for a 2,000-word article in one go. You keep more control and catch errors earlier.
• Rewrite introductions and conclusions yourself. These are where generic AI phrasing is most noticeable.
• Add something only you can provide in every piece: a client story (with permission), a photo, a worked example from your work, or a clear opinion.

For search visibility, remember that search engines and AI answer engines reward helpful, accurate content written for people. Our SEO, AEO and GEO guide covers this in more depth. Publishing large volumes of lightly edited AI text is a risky strategy and does not guarantee rankings or traffic.

Email and follow-up drafts

Routine emails are a good fit because you can check them in seconds. The AI email writer can draft a first version from a few bullet points, and the AI email reply generator can suggest a response to a message you paste in (after removing personal details you do not need to share).

Keep these rules:

• Never let an AI draft state a price, date, policy or promise you have not confirmed.
• Read every email before it goes out. Tone mistakes are easy to miss when skimming.
• For bulk marketing emails, the usual consent and unsubscribe rules still apply regardless of who or what wrote the copy.

Repurposing existing content

Repurposing is one of the highest-value uses because the source material is yours and already reviewed. A single webinar, podcast episode or long article can become:

• A short summary for your newsletter
• Three to five social posts, each built around one idea
• A list of frequently asked questions
• A short script for a video
• A checklist or one-page handout

The AI content repurposing tool is built for this. For platform-specific copy, the AI social media caption generator and AI hashtag generator can produce options to choose from. The AI YouTube title generator helps with title variations for video.

The main risk in repurposing is distortion: a summary that overstates a claim, drops a caveat or merges two separate points. Check each derived piece against the original.

Customer support drafts

AI can draft replies to common questions, but support is where accuracy and tone matter most. A safe setup:

1. Write a short, accurate knowledge base of your policies (returns, booking changes, delivery times, warranty terms). The AI FAQ generator can help structure it, but you must confirm every answer.
2. When drafting a reply, paste only the relevant policy text and the customer's question (without personal identifiers where possible).
3. Instruct the model to use only the policy text provided and to say "I'll check and get back to you" when the answer is not covered.
4. A person reviews, personalises and sends.

If you use a chatbot that replies to customers automatically, test it heavily, give customers a clear route to a human, and check any disclosure requirements that apply where you operate.

Meeting notes and internal operations

Turning a transcript or rough notes into a summary with decisions and action items saves real time. The AI meeting notes summarizer is designed for this. Always check the action items and owners against what was actually agreed, because summaries sometimes attribute tasks to the wrong person or invent deadlines.

Before recording or transcribing a meeting, tell participants and follow the consent rules that apply to call recording in your location.

Other operational uses include drafting standard operating procedures from a rough description, turning a checklist into a training document, and writing job descriptions (which you then check for fairness and legal compliance).

Analysis and reporting

AI can help you interpret a table, suggest ways to chart data, or draft a plain-language summary of monthly results. Three cautions:

• Check every number. Recalculate key figures in a spreadsheet or with our calculators, for example the percentage calculator or the ROAS calculator.
• Watch for invented explanations. A model may give a confident reason for a traffic drop that has no basis in your data. Treat explanations as hypotheses to test.
• Anonymise data first. Aggregate numbers (total leads per week) are usually safer to share than raw exports containing names, emails or phone numbers.

Product and service descriptions

The AI product description generator can turn a list of features into readable copy. Feed it your actual specifications and ask it not to add features that are not in the list. Then check every dimension, material, compatibility claim and care instruction. Incorrect product claims can create legal and returns problems.

A prompt-writing framework that works

Most disappointing AI output comes from vague prompts. A prompt is a brief, and like any brief it works better when it is specific. The framework below uses six parts, which you can remember as R-C-T-F-C-E: Role, Context, Task, Format, Constraints, Examples.

The six parts

1. Role. Who should the assistant act as? "You are an experienced copy editor for a small accounting firm." This sets tone and vocabulary.
2. Context. What does the assistant need to know? Audience, product, goal, brand voice, source material. This is the most important part and the one most often skipped.
3. Task. Exactly what you want. "Write three subject lines" is better than "help with my email."
4. Format. Length, structure, headings, bullet points, table, plain text, reading level.
5. Constraints. What to avoid: no invented facts, no prices, no exclamation marks, British or American spelling, words to avoid, claims that are not allowed in your sector.
6. Examples. A sample of the style you want, or a previous piece that performed well. Even one example improves consistency.

You do not need all six every time. For a quick rewrite, Task and Constraints may be enough. For anything customer-facing, include Context and Constraints at minimum.

The AI prompt generator can help you turn a rough idea into a structured prompt using this kind of structure.

Iterate instead of starting over

Treat the first response as a draft to steer. Useful follow-up instructions include:

• "Shorten this by about 30 percent without losing the three main points."
• "Make the tone warmer but keep it professional."
• "Which claims in this draft would need a source? List them."
• "Rewrite the second paragraph for someone who has never heard of this service."
• "Give me two alternative versions of the opening: one direct, one story-led."

Asking the model to list claims that need verification is a simple habit that makes your fact-checking faster.

Build a prompt library

Once a prompt produces good results, save it. A shared document or spreadsheet with columns for task, prompt text, when to use it, and notes on what to check is enough. Over time this becomes a team asset and keeps output consistent across people.

Reusable prompt templates

Copy these templates and replace the text in square brackets. Each includes a constraint asking the model not to invent facts, because that single instruction reduces (but does not remove) the risk.

Template 1: Blog outline

Role: You are a content strategist who writes practical guides for [audience, e.g. independent gym owners].
Context: I run [business description]. My readers are [who they are, what they already know, what they struggle with]. My brand voice is [e.g. plain, friendly, no jargon].
Task: Create an outline for an article titled "[working title]" that helps readers [specific outcome].
Format: H2 and H3 headings, with two or three bullet points under each describing what that section should cover. Include a section for common mistakes and a short checklist.
Constraints: Do not include statistics or claims that would need a source. Where a fact is needed, write [SOURCE NEEDED]. Avoid hype words.
My notes to include: [paste your own notes, examples, opinions].

Template 2: Email draft from bullet points

Role: You are writing on behalf of [name], [role] at [business].
Context: The recipient is [who they are and the relationship]. The purpose is [goal]. Relevant facts: [bullet list of confirmed facts only].
Task: Draft a short email.
Format: Subject line plus body under [120] words. One clear call to action: [action].
Constraints: Use only the facts listed. Do not mention prices, dates or commitments that are not in the list. Tone: [e.g. polite, direct, warm]. Spelling: [British/American].

Template 3: Repurpose one piece into several formats

Context: Below is an article I wrote for [audience]. [paste article]
Task: Create (1) a 150-word newsletter summary, (2) four social media posts for [platform], each focused on a different single idea from the article, and (3) five FAQs with short answers.
Constraints: Only use ideas and facts that appear in the article. Do not add new claims or statistics. Keep any caveats that the article includes. Flag anything you were unsure how to shorten accurately.
Format: Use clear headings for each output.

Template 4: Support reply using a policy

Role: You are a customer support assistant for [business].
Policy text (the only source you may use): [paste relevant policy]
Customer message: [paste message with personal details removed]
Task: Draft a reply that answers the question using only the policy text.
Constraints: If the policy does not cover the question, say we will check and reply within [timeframe]. Do not offer refunds, discounts or exceptions. Be warm and concise.

Template 5: Meeting summary

Task: Summarise the meeting notes below.
Format: (1) three-sentence summary, (2) decisions made, (3) action items as a table with columns: action, owner, due date. If an owner or due date was not stated, write "not stated" rather than guessing.
Constraints: Do not add information that is not in the notes.
Notes: [paste notes or transcript]

Template 6: Critique and improve your own draft

Role: You are a demanding editor.
Context: This is a [type of content] for [audience] with the goal of [goal].
Task: Review the draft below. List (1) unclear sentences, (2) claims that need evidence, (3) places where the reader might lose interest, (4) anything that sounds generic. Then suggest specific improvements. Do not rewrite the whole piece.
Draft: [paste]

Template 7: Data interpretation

Context: Below is a table of [metric] by [week/channel] for my [business type]. [paste aggregated table, no personal data]
Task: Describe the main patterns in plain language and suggest three hypotheses that could explain them, and what data I would need to test each one.
Constraints: Do not state causes as facts. Show any calculations you make so I can check them.

The human review workflow

Every piece of AI-assisted content should pass through a review before it reaches a customer or the public. The depth of review should match the risk. A social caption needs a quick read; a landing page with product claims needs a careful check.

Step 1: Fact-check

Go through the draft and mark every factual claim: numbers, dates, names, product specifications, legal or regulatory statements, quotes and references. For each one:

• Confirm it against a primary source (your own records, official documentation, the original study or regulator page).
• Remove it if you cannot confirm it.
• Replace vague claims with specific, verified ones where possible.

Be especially suspicious of neat statistics ("73 percent of customers...") and named studies. If you cannot find the original source, do not publish the claim.

Step 2: Brand voice and audience fit

Read the draft aloud. Does it sound like you? Common signs of unedited AI text include repetitive sentence structures, filler transitions, excessive hedging, lists of three everywhere, and a generic cheerful tone. Keep a short brand voice sheet (see the checklist below) and edit against it.

The AI grammar and tone assistant can suggest tone adjustments, but the final judgement is yours.

Step 3: Originality and rights

AI output can closely resemble existing material, and there is no guarantee that it is original. To reduce risk:

• Base drafts on your own notes and experience rather than asking for content "in the style of" a named competitor or author.
• Run important pieces through a plagiarism checker if you have access to one.
• Do not paste someone else's copyrighted article and ask for a rewrite to publish as your own. Paraphrasing tools, including the AI paraphrasing tool, are for improving your own text, not for disguising other people's work.
• Check the terms of any AI image tool regarding commercial use, and avoid generating images of real people or trademarked characters.

A note on "humanizing" tools: the AI text humanizer can make text read more naturally, but no tool can guarantee that content will be undetectable by AI detectors, and trying to disguise AI use where disclosure is required is a bad idea. Use such tools to improve readability, not to mislead.

Step 4: Disclosure where required

Disclosure rules for AI-generated or AI-assisted content vary by country, platform and sector, and they are changing. At the time of writing, some platforms ask creators to label realistic synthetic images or video, some advertising standards bodies expect disclosure where AI use could mislead consumers, and some sectors have their own rules. Check the current policies for each platform you publish on and any rules that apply to your industry. When in doubt, a short, honest disclosure is a low-cost choice.

Separately, never use AI to create fake reviews, fake testimonials, fake endorsements or images of customers who do not exist. That is misleading regardless of disclosure and may breach consumer protection law.

Step 5: Final approval

Decide who can approve what. A solo freelancer approves everything personally. A small team might agree that social posts need one reviewer, while web pages, ads and anything with prices or legal statements need the owner's sign-off.

Review checklist

• Every number, date, name and specification has been confirmed against a primary source
• Unverifiable claims and statistics have been removed
• No prices, policies or commitments appear that have not been confirmed
• The draft matches the brand voice sheet and has been read aloud
• The opening and closing have been rewritten or personalised by a person
• At least one element of first-hand experience or original insight has been added
• Content does not copy or closely paraphrase another source
• Images are properly licensed and do not depict real people without permission
• Disclosure requirements for the platform and sector have been checked
• The right person has approved the final version

Brand voice sheet (template)

We are: [three adjectives, e.g. clear, practical, friendly]
We are not: [three adjectives, e.g. salesy, jokey, academic]
Spelling and style: [e.g. British English, sentence case headings, no exclamation marks]
Words we use: [e.g. "clients" not "customers", "book" not "purchase"]
Words we avoid: [e.g. hype words, jargon list]
Claims we never make: [e.g. guaranteed results, medical claims]
Example of our voice: [paste a short paragraph you are happy with]

Paste this sheet into prompts as context. It improves first drafts and speeds up review.

Privacy and data handling

Many AI tools process your input on the provider's servers. Depending on the provider, plan and settings, inputs may be stored, reviewed by staff for safety, or used to improve models. These details differ between tools and change over time, so read the current terms and privacy settings for each tool you use.

The basic rule

Do not paste confidential or personal data into an AI tool until you have checked its terms, data retention settings and whether your use is permitted by your own obligations. This includes:

• Customer names, emails, phone numbers and addresses
• Health, financial or other sensitive personal information
• Contracts, client files and anything under a confidentiality agreement
• Passwords, API keys and access tokens (never share these)
• Unreleased financial results, pricing strategies or product plans
• Employee information

Practical safeguards

1. Minimise. Share only what the task needs. To draft a reply, you rarely need the customer's full name, order number and address.
2. Anonymise. Replace names with roles ("Customer A") and remove identifiers before pasting.
3. Aggregate. For analysis, use summary tables rather than raw exports.
4. Check settings. Many tools offer options to disable training on your inputs or to delete history. Business and enterprise plans often have different data terms from free plans. Confirm what applies to your account.
5. Write a short policy. Even a one-page AI use policy for your team (which tools are approved, what data may not be entered, who reviews output) prevents most problems.
6. Consider your obligations. If you handle personal data for clients, your contracts or applicable privacy laws may restrict sharing it with third-party processors. Check with a qualified professional if you are unsure. This guide is not legal advice.

AI use policy starter (template)

1. Approved tools: [list]. Other tools require approval from [name].
2. Never enter: personal data of customers or staff, passwords or keys, client confidential material, [other].
3. Allowed with anonymisation: support drafts, aggregated reports, [other].
4. All AI-assisted content is reviewed by a person before publishing or sending.
5. Factual claims must be verified against primary sources.
6. Disclosure: follow platform and sector rules; ask [name] if unsure.
7. Report mistakes or data concerns to [name] immediately.
Last reviewed: [date]

Limitations you need to plan around

Hallucinations

AI models can generate false information with complete confidence: made-up statistics, citations to studies that do not exist, incorrect product features, or fictional legal requirements. Asking the model "are you sure?" does not reliably fix this. The only dependable safeguard is verification against primary sources.

Warning signs include very precise numbers without a source, references you cannot find, and answers that perfectly match what you hoped to hear.

Outdated knowledge

Each model has a training cut-off, and its knowledge of platform features, prices, regulations and recent events may be out of date. Tools with web access can reduce this, but search results can themselves be wrong or outdated. For anything time-sensitive (ad platform settings, tax rules, privacy requirements), check the official source directly.

Bias and blind spots

Models reflect patterns in their training data, which can include stereotypes and gaps. This matters in audience personas, job ads, images and any content about groups of people. Review for assumptions about gender, age, culture, disability and location, and remember that a global audience may not share the default assumptions of the model.

Generic output

Because models produce likely text, unedited output tends to sound like everyone else's. This is a marketing problem as much as a quality one: generic content is hard to distinguish from competitors. The fix is your input: specific examples, opinions, customer language and first-hand detail.

Inconsistency

The same prompt can produce different answers each time. For repeatable tasks, use saved templates, include examples, and keep a human check in place.

Measuring time saved honestly

It is easy to feel faster with AI and still not be. The prompt, the regeneration, the editing and the fact-checking all take time. Measure before you assume.

A simple measurement method

1. Pick a repeatable task such as writing a weekly newsletter, drafting product descriptions or summarising client calls.
2. Record a baseline. Time the task three to five times without AI and take the average.
3. Time the AI-assisted version end to end, including writing the prompt, reviewing, fact-checking, editing and any rework after feedback.
4. Track quality alongside time. Use something observable: number of corrections needed, client revisions, email reply rate, or a simple 1 to 5 quality score from the reviewer.
5. Compare over several weeks, because the first attempts include learning time.

Worked example 1: a newsletter

These numbers are illustrative, not benchmarks.

• Baseline without AI: 90 minutes per newsletter.
• With AI: 20 minutes preparing notes and prompting, 35 minutes editing, 10 minutes fact-checking. Total: 20 + 35 + 10 = 65 minutes.
• Time saved per newsletter: 90 - 65 = 25 minutes.
• At four newsletters per month: 25 x 4 = 100 minutes saved, which is 1 hour 40 minutes.

That is useful but modest. If the quality score dropped or readers engaged less, the saving might not be worth it.

Worked example 2: product descriptions

Again, illustrative numbers.

• Baseline: 40 product descriptions at 15 minutes each = 40 x 15 = 600 minutes (10 hours).
• AI-assisted: 60 minutes to build and test a prompt template, then 6 minutes per product to generate, check specifications and edit. Total: 60 + (40 x 6) = 60 + 240 = 300 minutes (5 hours).
• Time saved: 600 - 300 = 300 minutes (5 hours).
• If your time is worth $50 per hour, that is 5 x $50 = $250 of time. If the tool costs $20 for the month, the net value is $250 - $20 = $230.

Notice that the setup time is included. For the next batch, the template already exists, so the saving would be larger. You can use the freelancer hourly rate calculator to estimate what an hour of your time is worth.

Time tracking table (template)

After a month, you will know which tasks genuinely benefit and which are faster by hand.

What to do with saved time

Time saved only matters if it goes somewhere useful. Decide in advance: more customer conversations, better offers, following up leads faster, or simply shorter working weeks. Otherwise the saving tends to disappear into more low-value content.

Building an AI-assisted weekly workflow

Here is an example weekly rhythm for a small business or freelancer. Adjust it to your channels.

Monday: plan

• Review last week's results (aggregated numbers only).
• Use AI to brainstorm three to five content ideas based on real customer questions from your inbox.
• Choose one core piece for the week.

Tuesday: create the core piece

• Write your own notes: key points, one story or example, your opinion.
• Use the outline template to structure it, then draft section by section.
• Run the full review workflow.

Wednesday: repurpose

• Use the repurposing template to create social posts, a newsletter summary and FAQs.
• Check each derived piece against the original.
• Resize and compress images for each platform with the social media image resizer and image compressor.

Thursday: communicate

• Draft follow-up emails and support templates for recurring questions.
• Summarise any client meetings and confirm action items.
• Add UTM parameters to links you share so you can see which channels drive visits, using the UTM link builder.

Friday: measure and improve

• Log time spent on AI-assisted tasks.
• Note which prompts worked and update the prompt library.
• Identify one process to improve next week.

Implementation checklist

• List five repeatable tasks and time each one manually
• Choose one or two approved AI tools and read their data terms and settings
• Write a one-page AI use policy and share it with anyone who helps you
• Create a brand voice sheet
• Save three prompt templates for your most common tasks
• Set up the review checklist as a standard step before publishing
• Track time and quality for four weeks
• Keep the tasks where AI saves time without lowering quality and drop the rest

Common mistakes

1. Publishing without review. The most common and most costly mistake. Even strong drafts contain errors.
2. Trusting statistics and citations from the model. Verify or remove them.
3. Pasting personal or confidential data without checking terms. Minimise, anonymise and check settings first.
4. Vague prompts. "Write a blog post about marketing" produces generic output. Give context, constraints and examples.
5. Asking for everything at once. Long single-shot outputs are harder to steer and review. Work in sections.
6. Ignoring brand voice. Without guidance, every business starts to sound the same.
7. Volume over value. Publishing more AI content is not a strategy. Fewer, better pieces with real insight usually serve readers and search visibility better, though no approach guarantees results.
8. Assuming AI knows current rules. Platform features, laws and prices change. Check official sources.
9. Using AI to fake social proof. Fake reviews, testimonials or endorsements are misleading and may be unlawful.
10. Not measuring. Without a baseline, you cannot tell whether AI is saving time.
11. Relying on detection or "undetectable" claims. No tool can guarantee content will pass or fail an AI detector. Focus on quality and honesty instead.

Glossary

• Large language model (LLM): A type of AI model trained on large amounts of text that generates text by predicting likely next words.
• Prompt: The instructions and context you give an AI assistant.
• Context window: The amount of text a model can consider at once, including your prompt and its response. Limits vary by tool.
• Hallucination: When an AI model produces false or invented information presented as fact.
• Training cut-off: The point after which a model has no built-in knowledge of events or changes.
• Temperature: A setting in some tools that controls how varied or predictable responses are.
• Few-shot prompting: Including one or more examples in the prompt to show the model the style or format you want.
• Repurposing: Turning one piece of content into several formats for different channels.
• Brand voice: The consistent personality and style of your written communication.
• Primary source: The original origin of a fact, such as official data, the original study, your own records or a regulator's page.
• Anonymisation: Removing or replacing details that identify a person before sharing data.
• Disclosure: Telling your audience that content was created or assisted by AI, where rules or honesty require it.

Next steps

1. Pick one repeatable task this week and time it manually as a baseline.
2. Use the AI prompt generator or the templates above to build a structured prompt for that task.
3. Run the AI-assisted version with the full review checklist and time it end to end.
4. Try the AI content repurposing tool on your best existing article or video.
5. Write your one-page AI use policy and brand voice sheet.
6. Read the Website Growth and Conversion Guide to make sure the traffic your content brings actually turns into enquiries.

Free toolAI Prompt GeneratorTurn a rough goal into a structured, reusable prompt for any AI assistant.

AI assistants are useful tools for drafting, transforming and organising work. They do not replace your knowledge of your customers, your judgement or your responsibility for what you publish. Keep people in the loop, protect your data, verify facts and measure honestly, and AI can become a dependable part of how you work.