Tips & tricks

Get more from AI, with less guesswork

Practical tips from the AI companies' own guidance, how to check what AI tells you, a guide to AI agents, and a prompt builder for 28 tools. Every tip links to its source.

Prompt builder

Free. We don't store what you type.

Get better answers

What OpenAI, Anthropic, Google and Microsoft recommend, in plain English.

Say exactly what you want and why

Start with a clear action, such as draft, plan, summarise or compare, then add who it is for and what you will use it for. OpenAI suggests using an action verb and being clear about the task, and Microsoft says a clear goal is the one part of a prompt you always need.

Works with: Any chat assistant

See an example

Instead of

Write something about our new opening hours.

Try

Draft a short, friendly email to our regular customers announcing that from 1 November we open at 8am instead of 9am on weekdays. Keep it under 120 words.

Sources: openai.com · support.microsoft.com

Give the background a newcomer would need

Write your prompt as if the AI knows nothing about your situation. Anthropic advises giving 'as much context as possible' and explaining why you want something, because the reason helps the model understand your goal. Google gives the same advice for Gemini.

Works with: Any chat assistant

See an example

Instead of

Make this CV better.

Try

I'm applying for a junior data analyst role at a local council. I've just finished an apprenticeship and have no degree. Here's my CV. Suggest changes that make my Excel and SQL experience stand out to that kind of employer.

Sources: support.claude.com · platform.claude.com · support.google.com

Show an example of what you want

If you have a style or layout in mind, paste a sample rather than describing it. Anthropic calls examples 'one of the most reliable ways' to steer format, tone and structure, and OpenAI and Google both recommend showing the format through examples.

Works with: Any chat assistant

See an example

Instead of

Write product descriptions for my candles.

Try

Write descriptions for these three candles in the same style as this one: 'Sea Salt & Sage. Fresh, clean and calming. Hand-poured soy wax, 40-hour burn.' Candles: Fig & Cedar, Lemon Verbena, Smoked Vanilla.

Sources: platform.claude.com · help.openai.com · ai.google.dev

Set the format, length and tone

Say how you want the answer laid out: a table, bullet points, a short paragraph, or a step-by-step list. Be precise about length, for example '3 to 5 sentences' rather than 'fairly short', and name the tone, such as formal, friendly or plain English.

Works with: Any chat assistant

See an example

Instead of

Explain pensions to me, keep it short.

Try

Explain how a UK workplace pension works in plain English, in 5 bullet points of one sentence each, for someone starting their first job.

Sources: help.openai.com · help.openai.com · ai.google.dev

Say what to do, not what to avoid

OpenAI advises that 'instead of just saying what not to do, say what to do instead'. A positive instruction gives the AI a clear target, while a list of bans leaves it guessing what you actually want.

Works with: Any chat assistant

See an example

Instead of

Don't use jargon.

Try

Use everyday words a 12-year-old would understand, and explain any technical term in brackets the first time you use it.

Source: help.openai.com

Break big jobs into smaller steps

If your request has several parts, ask for them one at a time or in a numbered order. Anthropic, Google and OpenAI all recommend splitting complex requests into smaller steps.

Works with: Any chat assistant

See an example

Instead of

Plan my wedding.

Try

First, list the main jobs for a 60-guest wedding next June, grouped by month. Once I've checked that, we'll work on the budget.

Sources: support.claude.com · support.google.com · openai.com

Refine the answer with follow-up messages

Treat the first reply as a draft and reply with what to change, such as 'shorter', 'more formal' or 'add a section on costs', rather than starting again. OpenAI, Microsoft and Anthropic all recommend following up to refine an answer. When you switch to a different task, Microsoft suggests typing 'new topic'.

Works with: Any chat assistant

See an example

Try

Good start. Cut it to half the length, drop the second paragraph, and make the ending a clear call to action.

Sources: help.openai.com · support.microsoft.com · support.claude.com

Ask for options and say what matters

If you want choices, ask for them, for example 'give me three versions'. OpenAI also suggests telling ChatGPT what matters most to you, such as accuracy, creativity or speed, so it knows what to trade off.

Works with: Any chat assistant

See an example

Instead of

Give me a name for my dog-walking business.

Try

Suggest five names for a dog-walking business in Leeds. Easy to say and spell matters most; clever puns matter least.

Source: openai.com

Ask it to quote and check itself

Anthropic's guidance suggests ending a prompt with an instruction such as 'Before you finish, verify your answer against...' to catch errors, and asking the model to quote the relevant parts of a long document before answering. This is written for Claude but is easy to try with other assistants. It reduces mistakes but does not replace your own checking.

Works with: Claude (official guidance); worth trying with any assistant

See an example

Instead of

What does this tenancy agreement say about ending the contract early?

Try

Quote the exact clauses in this tenancy agreement about ending the contract early, then explain them in plain English. Before you finish, check every point you make against the quoted text.

Source: platform.claude.com

Attach the document instead of describing it

Upload the actual file so the AI can summarise, pull out details or combine information from it; OpenAI recommends adding files, images and documents as context. On every ChatGPT plan except Enterprise, only the text in uploaded documents is read and images inside them are discarded. With very long text, Anthropic advises putting the document first and your question after it.

Works with: ChatGPT, Claude, Gemini and Copilot (file support varies by app and plan)

See an example

Try

[attach meeting-notes.pdf] List every action point in these notes with the person responsible and the deadline, as a table.

Sources: help.openai.com · openai.com · platform.claude.com

Save your preferences with custom instructions

Instead of repeating 'I'm in the UK, use British spelling' in every chat, put it in ChatGPT's custom instructions (Settings > Personalization on web and desktop). They are available on all plans and apply to all your chats. ChatGPT's Memory can also remember details between chats, and you can manage or turn it off under Settings > Personalization > Memory.

Works with: ChatGPT

See an example

Try

Custom instructions: I live in the UK. Use British spelling, pounds and metric units. Keep answers short unless I ask for more detail.

Sources: help.openai.com · help.openai.com

Pick the right mode for the job

Anthropic suggests web search for quick factual questions and recent news, extended thinking for complex reasoning that doesn't need the web, and Research for in-depth information gathering. ChatGPT's deep research is meant for multi-step questions that combine many sources and produces a cited report, while OpenAI says standard chat may be faster for quick lookups. OpenAI notes deep research availability depends on your plan and country.

Works with: Claude, ChatGPT

Sources: support.claude.com · help.openai.com

Check answers and stay safe

AI can be confidently wrong. A few habits catch most problems.

Remember AI can be confidently wrong

OpenAI says ChatGPT 'may express high confidence even in incorrect answers', and Anthropic says Claude can 'hallucinate' incorrect or misleading information. The UK's NCSC warns that answers can sound convincing but be only partly correct, especially on niche topics. Microsoft adds that the same prompt can give different answers each time.

Works with: Any chat assistant

Sources: help.openai.com · support.claude.com · ncsc.gov.uk · support.microsoft.com

Check facts, figures and sources yourself

AI can invent quotes, studies and references to sources that don't exist. OpenAI advises you to 'always verify quotes, data, technical information or references', and Anthropic suggests opening the cited websites because the original may contain context the summary missed. Treat AI output as a first draft and cross-check anything important with a trusted source.

Works with: Any chat assistant

Sources: help.openai.com · support.claude.com · support.microsoft.com

Keep personal and confidential details out

The NCSC advises not to put sensitive information into public AI chatbots, or anything that would cause problems if it were made public. Google asks you not to enter confidential information you wouldn't want a human reviewer to see, as some Gemini chats are reviewed by people and reviewed chats can be kept for up to three years. Remove names, account numbers and work secrets before you paste.

Works with: Any chat assistant

Sources: ncsc.gov.uk · support.google.com

Decide whether your chats train the AI

Each service has its own setting: in ChatGPT, go to Settings > Data controls and turn off 'Improve the model for everyone'; in Claude, use the 'Model Improvement' option in Privacy Settings. In Gemini, turning off 'Keep Activity' stops future chats being used for training, unless you choose to send Google feedback. In Copilot, go to Privacy and turn off 'Training on conversation activity' and 'Training on voice conversations'.

Works with: ChatGPT, Claude, Gemini, Copilot

Sources: help.openai.com · privacy.claude.com · support.google.com · support.microsoft.com

Use temporary or incognito chats for one-offs

ChatGPT's Temporary Chat doesn't appear in your history, isn't used to train models and doesn't create memories. Claude's Incognito chats aren't used for training even if Model Improvement is on, and Gemini's temporary chats aren't used to train Google's models. This doesn't make them a safe place for very sensitive information.

Works with: ChatGPT, Claude, Gemini

Sources: help.openai.com · help.openai.com · privacy.claude.com · support.google.com

Get a professional for legal, medical, money decisions

Anthropic says users 'should not rely on Claude as a singular source of truth' and should carefully scrutinise high-stakes advice. OpenAI's usage policies rule out tailored legal or medical advice without a licensed professional involved. Use AI to understand the basics and prepare questions, then check with a qualified person; for investments, Surrey's Police and Crime Commissioner advises checking the firm with the FCA.

Works with: Any chat assistant

Sources: support.claude.com · openai.com · surrey-pcc.gov.uk

Watch for AI-powered scams and deepfakes

Surrey's Police and Crime Commissioner warns that criminals can clone a voice from as little as three seconds of audio and use fake videos of celebrities, and suggests calling back on a number you know and agreeing a family code word. The NCSC's warning signs still apply: urgency, strong emotion and pressure to act without thinking. Forward scam texts to 7726 and report scam emails through the NCSC.

Works with: Everyone

Sources: surrey-pcc.gov.uk · ncsc.gov.uk · ncsc.gov.uk · ncsc.gov.uk

AI agents

An AI agent is an AI system that does not just answer you, it carries out a task. IBM describes an agent as a system that autonomously performs tasks by designing workflows with the tools it has. Anthropic puts it as a system where the AI model directs its own process and decides which tools to use, rather than following a fixed script.

Full agents guide

Give it one clear goal and a finish line

Say exactly what 'done' looks like. OpenAI warns against vague requests such as 'check my email and handle everything', and advises giving agents specific, narrow tasks rather than broad freedom. A tight brief also gives hidden instructions on web pages less room to steer the agent.

See an example
Find three train times from Leeds to London on Friday morning arriving before 11am, and list them with prices. Do not book anything.

Sources: help.openai.com · openai.com

Set your limits up front

Tell the agent what it must not do and what its boundaries are: budget, dates, which sites to use, and that it should stop and ask before anything that costs money or contacts someone. Specific instructions are one of OpenAI's four recommended protections against manipulation.

See an example
Budget is £150 maximum. Only use the hotel's own website or Booking.com. Stop and ask me before entering any payment details.

Sources: openai.com · support.microsoft.com

Only connect what the job needs

Switch off apps and connectors the task does not need, and review app permissions regularly. The UK's ICO flags agents processing personal information beyond what is necessary as a key data protection risk.

Sources: help.openai.com · ico.org.uk

Which agents to use, how to brief them and what can go wrong

Writing & chat

ChatGPT

Help me write an email asking my landlord to fix the heating

Write a polite but firm email to my landlord asking them to repair the heating.

Context: The heating in my flat has not worked for 5 days. I reported it by phone on Monday and have had no reply. It is getting cold and I work from home.

Requirements:
- 120-180 words, short paragraphs, main request in the first two sentences
- Ask for a repair date within 3 working days and a reply by Friday
- Courteous, factual tone; no threats
- Include a clear subject line

Use [square-bracket placeholders] for my name, address and the landlord's name.
How to prompt ChatGPT
Claude

Explain inflation to my 12 year old

<context>
I want to explain inflation to my 12-year-old, who is curious and reads well but has no economics background. The goal is for them to understand why prices go up over time and why it matters for pocket money and savings.
</context>

<instructions>
Explain inflation in plain language using one everyday example (such as the price of a favourite snack over several years). Then explain one cause of inflation and one effect on savings. Finish with two questions I can ask them to check they understood.
</instructions>

<format>
Write 200-250 words in short, friendly paragraphs suitable to read aloud. Avoid jargon; if you must use a term, define it in the same sentence.
</format>
How to prompt Claude
Google Gemini

Plan a 3 day trip to rome

You are an experienced travel planner.

Task: Create a 3-day itinerary for Rome.

Context: Two adults, first visit, mid-range budget, staying near Termini station, travelling in late October. We enjoy history and food and prefer walking or public transport.

Constraints:
- Group sights by neighbourhood to limit travel time
- Include booking tips for sights that need tickets
- Suggest one lunch and one dinner spot per day

Format: For each day, a table with Time, Activity, Notes. End with 5 bullet points of practical tips. Use Google Search to check current opening hours and note anything uncertain.
How to prompt Google Gemini
Microsoft Copilot

Summarise this report for my boss

Goal: Summarise the attached quarterly sales report for my manager.

Context: My manager has 2 minutes to read this before a leadership meeting and cares most about revenue against target, the biggest risks, and decisions needed.

Source: Use only the attached report "Q3 Sales Review".

Expectations: An executive summary of no more than 150 words, followed by 3 bullet points for risks and 2 bullet points for decisions needed. Professional, neutral tone. Flag any figures that look inconsistent.
How to prompt Microsoft Copilot

Images

Midjourney

A cozy coffee shop poster for autumn

Vintage screen-printed poster of a small corner coffee shop on a rainy autumn evening, warm amber light glowing from tall windows, maple leaves drifting past a striped awning, a steaming mug in the foreground, burnt orange, mustard and deep teal palette, bold headline "AUTUMN BREWS" in chunky cream serif letters across the top --ar 2:3 --s 150 --no people
How to prompt Midjourney
ChatGPT Images (OpenAI gpt-image-2)

A cozy coffee shop poster for autumn

Purpose: a printable vertical poster (2:3) promoting a coffee shop's autumn menu.
Scene: a small corner café on a rainy autumn evening, maple leaves on the wet pavement, warm amber light spilling from tall windows.
Subject: a steaming ceramic mug of spiced latte with latte art in the foreground on a wooden window ledge.
Style: flat, vintage screen-print illustration with visible paper grain; palette of burnt orange, mustard yellow, deep teal and cream.
Text: headline "AUTUMN BREWS" in bold cream serif capitals across the top; smaller line "Pumpkin spice • Maple latte • Chai" in a clean sans-serif along the bottom.
Constraints: no people, no extra text or logos, keep the composition uncluttered.
How to prompt ChatGPT Images (OpenAI gpt-image-2)
Google Gemini (Nano Banana 2 / Nano Banana Pro)

A cozy coffee shop poster for autumn

Create a vertical 2:3 poster in a vintage screen-printed illustration style for a neighbourhood coffee shop's autumn menu. A steaming ceramic mug of spiced latte sits on a wooden window ledge in the foreground; behind it, a rainy street at dusk with maple leaves scattered on the wet pavement and warm amber light glowing from the café's tall windows. Soft, cozy atmosphere with a burnt orange, mustard and deep teal palette and visible paper grain. The headline "AUTUMN BREWS" runs across the top in bold cream serif capitals, and the line "Pumpkin spice • Maple latte • Chai" sits along the bottom in a clean sans-serif. The street is quiet and empty.
How to prompt Google Gemini (Nano Banana 2 / Nano Banana Pro)
Ideogram

A cozy coffee shop poster for autumn

A vintage screen-printed poster with the headline "AUTUMN BREWS" in bold cream serif capitals across the top and "Pumpkin spice • Maple latte • Chai" in small clean sans-serif along the bottom. A steaming ceramic mug of spiced latte sits on a wooden window ledge in the foreground. Behind it, a quiet, empty street on a rainy autumn evening, maple leaves scattered on wet pavement. Warm amber light glows from tall café windows. Centered vertical composition with plenty of space around the text. Flat illustration with visible paper grain in burnt orange, mustard and deep teal.
How to prompt Ideogram

Video

Runway

A cowboy on a horse at sunset

Low-angle tracking shot of a weathered cowboy on a rearing black horse on a dusty ridge at sunset. Strong backlight from the low sun creates a golden rim light and silhouettes of dust in the air. [00:00 through 00:03] The horse rears up and the cowboy grips the reins. [00:03 through 00:06] The horse lands and gallops toward camera as the camera slowly pulls back. Classic 35mm western film look, warm orange palette.
How to prompt Runway
Kling AI

A couple reacting to a plot twist on TV

Cozy living room at night lit only by the flickering blue glow of a TV. A middle-aged couple sits on a beige sofa sharing a bowl of popcorn. Shot 1, medium two-shot, static camera: both stare at the screen and freeze mid-bite. Shot 2, slow push-in on the woman's face: her eyes widen. Mom (softly, in a surprised tone, British English): "Wow, I did not see that coming." Shot 3, close-up of the man, handheld: Dad (low voice, calm, agreeing): "Yeah… nobody did." Ambient sound: muffled TV dialogue, popcorn crunch. Warm, intimate, cinematic.
How to prompt Kling AI
Google Veo / Gemini Omni (Gemini app & Flow)

A red convertible driving along the coast

Drone shot following a classic red convertible along a winding coastal cliff road at golden hour, waves crashing on the rocks below. One continuous shot. Warm, low sunlight with long shadows; cinematic 35mm film look. The driver, a woman in her thirties with a white headscarf and sunglasses, glances at the sea and smiles. Audio: the engine roars as it accelerates, wind rushing, distant gulls, surf crashing. No music.
How to prompt Google Veo / Gemini Omni (Gemini app & Flow)
Luma (Dream Machine / Ray)

A runner in Tokyo at night

Slow tracking shot alongside a female runner in a black windbreaker on a quiet Tokyo backstreet at night. Wet asphalt mirrors soft pink and cyan neon signs; light rain falls through the glow of a single streetlamp. Shallow depth of field, anamorphic lens flares, teal-and-magenta palette, cinematic 35mm film grain.
How to prompt Luma (Dream Machine / Ray)

Music & voice

Suno

A sad country song about a dog

STYLES: modern country ballad, slow tempo ~72 BPM, fingerpicked acoustic guitar, pedal steel, soft piano, warm male baritone vocals, intimate live-room production, nostalgic and bittersweet
EXCLUDE: electronic drums, autotune
LYRICS:
[Intro: pedal steel]
[Verse 1]
There's a dent in the porch where you used to lie
A red ball fading in the July sun
[Chorus]
Old friend, I still leave the gate open wide
In case you're running home tonight
[Verse 2]
Your collar hangs on the truck mirror still
It rattles every time I drive past the hill
[Bridge: whispered vocals, guitar only]
Good boy… good boy
[Chorus: full band]
Old friend, I still leave the gate open wide
In case you're running home tonight
[Outro]
[End]
How to prompt Suno
ElevenLabs

An excited announcement that the cafe is reopening

[excited] Guess what? After three long months… we're BACK! [laughs] The doors open this Saturday at eight. [whispers] And the first coffee… is on us. [cheerfully] See you there!
How to prompt ElevenLabs

Apps & code

Lovable

An app for my dog walking business

Build a booking web app for a solo dog-walking business. Users: dog owners booking walks on their phones, and me (the walker) managing my schedule.

Core journey: owner signs up -> adds dog profile (name, breed, notes) -> picks a 30 or 60 min slot from my availability -> gets confirmation; I see today's walks in order.

Start with just the public landing page and the booking flow (no admin yet).

Visual style: friendly and playful but tidy: warm off-white background, forest-green primary, rounded cards, a rounded sans-serif, generous spacing. Use real copy: headline 'Happy dogs, tired legs', 3 benefit cards, 2 pricing cards (30 min £12, 60 min £20).

States: loading skeletons, an empty state when no slots are free, inline form errors.
Responsive: mobile-first; booking steps stack on mobile, two columns on desktop.
Data: user accounts with email login, tables for dogs, bookings, availability.

Ask me any questions you need to fully understand what I want before building.
How to prompt Lovable
Bolt.new

A habit tracker app

Build a habit tracker web app for individuals who want to build 3-5 daily habits.

Stack: React + Vite + TypeScript, Tailwind CSS, Supabase for auth and data.

Step 1 only: project setup, email/password login, and a 'Today' screen listing my habits with a checkbox each. Don't build stats or reminders yet.

Data: habits (id, user_id, name, colour, created_at), completions (habit_id, date).

UI: single-column mobile-first layout, white cards on a light grey background, one accent colour per habit, large tap targets, an empty state with an 'Add your first habit' button.

Add clear error messages and console logging around all Supabase calls.
How to prompt Bolt.new
v0

A pricing page for my SaaS

Create a pricing page for a B2B SaaS that schedules social media posts, aimed at small marketing teams.

Stack: Next.js App Router, Tailwind, shadcn/ui components.

Layout: headline + subhead, monthly/annual toggle (annual shows '2 months free'), 3 tier cards (Starter $19, Team $49 highlighted as 'Most popular', Agency $129) with 5 feature bullets each, a feature comparison table, and a 5-question FAQ accordion.

Style: clean and confident: white background, indigo accent, Inter, soft shadows, rounded-xl cards.

Behaviour: toggle updates prices without reload; CTA buttons link to /signup?plan=<tier>. Fully responsive: cards stack on mobile.

If anything is unclear, ask me before building.
How to prompt v0
Replit Agent

An expense tracker for my team

Build a web app where a small team (5-15 people) logs work expenses and a manager approves them.

User flow: employee logs in -> submits expense (date, amount in GBP, category, description, receipt photo) -> manager sees a pending list -> approves or rejects with a comment -> employee sees the status.

Stage 1 only: login, the submit form, and the employee's own expense list. Approval comes next.

Requirements:
- Amount must be a positive number with 2 decimals
- Categories: Travel, Meals, Equipment, Other
- Receipt upload accepts JPG/PNG/PDF up to 5 MB
- List sorted newest first, with status badges (Pending/Approved/Rejected)

Use a PostgreSQL database. Plan the data model first and show me the task list before building.
How to prompt Replit Agent

Presentations

Gamma

A presentation about our Q3 results

Create a 10-slide presentation of Q3 results for our company leadership team (non-finance execs) so they can approve the Q4 hiring plan.

Tone: confident, plain-English, no jargon. Text: brief, max 3 bullets per slide, one key message per slide title.

Outline:
1. Title: Q3 2026 Results
2. Headline: revenue £4.2M (+18% QoQ)
3. Revenue by product line (chart)
4. Customer growth: 1,240 -> 1,510 accounts
5. Churn down 3.1% -> 2.4%
6. What worked
7. What didn't
8. Q4 priorities
9. Hiring ask: 6 roles, £420k
10. Decision needed + next steps

Visuals: clean corporate style, navy and teal, simple charts and icons, no stock photos of people.
How to prompt Gamma

The Weekly Tally

Useful updates, insights and no-hype advice. Read this week's digest.

One email a week. Unsubscribe in one click.