This morning, I asked an AI to summarise my unread emails while I made coffee. Then I used Perplexity to fact-check something before sending an article to my editor. Then Remove.bg erased the background from a photo in four seconds — a task that used to take me twenty minutes in Photoshop. And on the way out, I asked my phone's voice assistant to remind me to buy milk at 6pm.
That's when it hit me, zipping up my jacket: when exactly did I start doing all of this without thinking? None of those four actions would have been possible five years ago — or at least, not at this speed, this ease. I've become an office cyborg without noticing, and I'm not sure I fully understand what that means.
This guide is the result of several months of genuinely testing these tools — not just reading their landing pages. I've included what works, what's overhyped, what things actually cost, and the ethical questions you'd be wise to think about before delegating everything to a language model. I'm Kristina, I'm passionate about technology, and I'm not trying to sell you anything.
Ready to separate the genuinely useful from the hype? Let's go.
AI for writing and communication: what actually works
Let's start with the most crowded sector in the market. You've been sold AI writing tools since 2022, and honestly, quality varies from "genuinely impressive" to "catastrophically generic". Here's what I've actually tested.
ChatGPT (OpenAI) — The indispensable generalist
Price: Free (GPT-3.5) / £16/month for ChatGPT Plus (GPT-4o). The free version is largely usable, but limited in context window and access to advanced features.
ChatGPT is most people's entry point, and for good reason: it's accessible, multilingual, and capable of handling very diverse requests. For everyday writing — drafting a tricky email, reformulating a paragraph, generating outline ideas — it's excellent. It's not magic; it's a very good assistant.
What works well with ChatGPT:
- Improving an existing draft: paste your text and ask it to "tighten it up" or make it "more professional" — good results in 30 seconds.
- Changing register: converting a formal email into a casual message, or vice versa.
- Generating variations: 5 different email subject lines, 3 ways to approach a difficult topic.
- Long summaries: paste an article or report and ask for the key points as 5 bullets.
- Nuanced translation: not just word-for-word translation, but "how would you naturally say this in British English in a professional context".
What works much less well:
- Highly specialised content: GPT-4o can produce factual errors on technical niche subjects with disconcerting confidence. Always verify data and statistics.
- Very personal voice: text generated without precise instructions sounds like ChatGPT, not like you. You need to train the model with examples of your own writing.
- Freshness of information: the free version has a knowledge cutoff. ChatGPT Plus with browsing enabled is more reliable, but still less good than Perplexity for recent factual research.
Claude (Anthropic) — My main tool for long-form writing
Price: Free (limited) / Claude Pro at £14/month. The free Claude 3 Haiku model is surprisingly capable for basic tasks.
I'll be direct: for long-form writing and stylistic consistency over several thousand words, Claude is currently better than ChatGPT. Its context window (200K tokens on Pro versions) allows it to ingest an entire document and work on it consistently. It follows complex instructions better and produces less "mechanical" text in creative registers.
My concrete use: I give it a detailed brief with examples of my style, and it produces a first draft that I reshape rather than starting from a blank page. The time saving is real — around 40% on in-depth articles, based on my personal measurements over six months.
Never ask an AI to "write an article about X". Give it a precise role, a target audience, a tone, length constraints, and paste examples of what you like. The quality of the output is directly proportional to the quality of your prompt. A good prompt takes 5 minutes to write and is the difference between generic content and something actually usable.
Notion AI — The integrated assistant within your workspace
Price: £7/month as a Notion AI add-on (on top of the Notion subscription). Included in Notion Plus at £12/month.
Notion AI is relevant if you're already in the Notion ecosystem for project management or note-taking. The integration is seamless: you select a text block and ask it to "improve", "summarise", "continue", without leaving your workspace. For heavy Notion users, this is a genuine friction reduction.
The main limitation: the model's intrinsic quality is below GPT-4o or Claude for complex tasks. For serious writing, I always return to Claude. Notion AI is the comfort tool for quick touchups within the workflow.
Grammarly — The correction assistant that became an AI
Price: Free (basic correction) / Premium at £9/month / Business at £12/user/month.
Grammarly is the veteran that added generative AI to its correction engine. The result is a useful tool for English writers — grammar and style checking remains its main strength, and it's excellent. The generative assistant is decent but not exceptional compared to ChatGPT or Claude.
My honest take: if you write primarily in English, the free version is already useful for proofreading. The Premium version is justified if you produce a lot of professional English content. For other use cases, there are better alternatives for the money.
AI image generation: what for, and at what cost
AI image generation is the sector that most shocked the public, sparked the most ethical debates, and produced the most unjustified hype. Let's take this seriously.
Midjourney — The most impressive artistically
Price: No free subscription since early 2024. Basic Plan at $10/month (approx. 200 images/month), Standard Plan at $30/month (unlimited images in "relax mode"). Works exclusively via Discord.
Midjourney produces the most aesthetically accomplished images on the market for conceptual photography, illustration, and artistic visuals. v6 (2024) generates correct hands, which was historically the number-one problem. Understanding of complex prompts is excellent.
Use cases where Midjourney adds genuine value:
- Social media visuals if you have a precise graphic style to explore
- Moodboards and visual research for creative projects
- Editorial illustrations (with the rights questions this raises — see the limitations section)
- Presentation images that don't require identifiable real people
Midjourney is not designed for:
- Retouching or modifying an existing photo (it's a pure creation tool)
- Logos or visual identities (text generation remains problematic despite progress)
- Anyone who wants an intuitive interface — Discord remains a barrier to mainstream adoption
DALL-E 3 (integrated in ChatGPT Plus) — The accessible option
Price: Included in ChatGPT Plus (£16/month). Also available via the OpenAI API per generation.
DALL-E 3 has a unique advantage: it's directly integrated into ChatGPT Plus, allowing you to generate images within a conversation. You can describe what you want in natural language, iterate by chatting, and get results without leaving the interface. It's less impressive artistically than Midjourney v6, but much more accessible.
It's my default tool for quick cover images, blog illustrations, and conceptual diagrams. Quality is sufficient for 80% of digital needs, and the conversational integration is a genuine advantage.
Canva AI — AI for non-designers
Price: Free Canva includes some AI credits. Canva Pro at £10/month unlocks the full Magic Studio.
Canva has integrated generative AI into its design offering under the name Magic Studio. Magic Media generates images, Magic Write assists with writing, and Magic Eraser removes elements from a photo. For someone already using Canva to create visuals, it's a natural extension.
Canva's AI isn't the most powerful on the market, but it's the best contextualised within a graphic design workflow. For a small business or freelancer managing their communications solo, it's probably the most practical tool of the three.
The question of copyright over AI-generated images is still being resolved legally in the UK and internationally. Under current UK law, AI-generated images may not attract copyright protection where there is no qualifying human creative contribution. This also means that the AI images you generate may not "belong" to you in any strict legal sense — check the terms of service of each platform carefully. Midjourney and DALL-E have different commercial licensing conditions. And ethically: were the training datasets compiled using artists' work without consent or compensation? The answer is generally yes, and several lawsuits are ongoing in the UK and US.
Productivity and organisation: tools with measurable time savings
This is where AI genuinely delivers on its promises — provided you choose the right tools for the right use cases. I've measured my time savings over six months. The results are mixed but real.
Otter.ai — Meeting transcription that changes everything
Price: Free (600 minutes/month) / Pro at $10/month (unlimited) / Business at $20/user/month.
Otter.ai transcribes your meetings in real time, identifies speakers, and generates an automatic summary with action points. Since I started using it for all my client meetings, I've barely taken notes during conversations — which allows me to be fully present in the exchange.
Measured time saving: approximately 45 minutes per week on average, between taking notes during the meeting and writing up the summary afterwards. Over a year, that's around 38 hours recovered. At $10/month, the ROI is obvious.
Important limitations:
- Transcription accuracy drops noticeably for strong regional accents, specialised technical terms, and multi-speaker conversations.
- The app requires microphone access — which raises confidentiality questions for sensitive meetings.
- Automatic summaries are drafts, not finished deliverables. They require proofreading and reformulation.
Notion AI — Beyond writing
Beyond writing assistance, Notion AI allows you to query your personal knowledge base. If you store your notes, articles, and resources in Notion, you can ask questions in natural language and get syntheses in response. That's the promise of the accessible "second brain". In practice, it works well if your Notion database is well-structured and kept up to date — which is itself a significant discipline effort.
Reclaim AI — Automatic calendar scheduling
Price: Free (basic features) / Starter at $8/month / Business at $12/user/month.
Reclaim AI connects to your Google Calendar and automates the scheduling of your recurring tasks (exercise, deep work, breaks), your daily priorities, and your habits. It understands your constraints and automatically suggests optimal slots. If your week looks like a Tetris game between meetings and to-dos, this is life-changing.
What I observed after three months of use: fewer micro-decisions about "when am I going to do that", and better consistency between what I plan to do and what I actually do. It's not magic — the tool doesn't replace discipline — but it reduces the cognitive load of scheduling.
Motion — The more powerful alternative
Price: $19/month (individual) / $12/user/month (teams).
Motion is more powerful than Reclaim in its integration of to-do list and calendar: it takes your tasks, deadlines, and availability, and automatically builds your daily schedule. When a meeting is added unexpectedly, it readjusts the entire schedule in real time.
It's impressive in demos. In reality, the learning curve is steep, and if you don't feed the system with real deadlines and time estimates, the suggestions aren't relevant. It's a powerful tool for people who already have good task management discipline and want to automate it — not for those looking to get organised for the first time.
Start with one AI productivity tool and genuinely integrate it into your workflow for 30 days before adding another. The classic mistake: signing up for 5 tools in one week, using each for 3 days, and concluding that "AI is useless". Otter.ai or Reclaim alone, well used, is worth more than Motion + Notion AI + 3 other tools used superficially. Depth beats breadth.
Research and information: Perplexity vs Google, Claude vs ChatGPT
This is the area where AI has changed my way of working the most — and also where the risks of misinformation are most serious. I can no longer pretend Google is my primary research tool since I discovered Perplexity.
Perplexity AI — My go-to search engine
Price: Free (very generous) / Pro at $20/month for the most powerful models and unlimited searches.
Perplexity combines real-time web search with a language model. In concrete terms: you ask a question in natural language, Perplexity searches the web, synthesises sources, and presents a structured answer with clickable citations. It's accelerated research, not hallucinatory generation.
The fundamental difference from ChatGPT or Claude: Perplexity cites its sources explicitly and systematically. For each point in the response, you can see where the information comes from. It's verifiable. That's the minimum condition for me to trust a research tool.
Use cases where Perplexity outperforms Google:
- Questions requiring synthesis: "What are the differences between Notion, Coda, and Obsidian subscription models?" — Google gives you 10 links to compare, Perplexity gives you the answer.
- Technical comparisons: "Claude vs ChatGPT for long-form writing in 2024" — synthesis in 60 seconds with sources.
- Sector monitoring: latest news on a topic with context.
- Recent factual questions: statistics, recently published studies.
Cases where Google is still better:
- Local searches (restaurants, addresses, opening times)
- Navigating to a specific site whose URL you don't remember
- Images, videos, shopping
- Very specific queries requiring a precise source rather than a synthesis
Even Perplexity can be wrong and present incorrect information with corresponding sources. I've had cases where the synthesis was erroneous but the source links were correct — the model had misinterpreted the sources. For any information you're going to reuse (in an article, a presentation, an important email), click on the sources and read them yourself. Perplexity accelerates research; it doesn't replace critical thinking. The Ada Lovelace Institute's 2024 report on AI and information integrity makes exactly this point: the appearance of citation is not the same as verified accuracy.
Claude for document analysis
An often-underestimated use of Claude: submitting documents for analysis. Annual report PDFs, contracts to review, dense academic papers — Claude can extract key points, identify contradictions, compare two documents, or answer specific questions about a body of text. Its 200K token context window allows it to ingest very long documents without losing the thread.
For freelancers, lawyers, researchers, or anyone who works with lots of text documents, this is a considerable time saving on reading and synthesis tasks. I timed it: analysing an 80-page report takes 4 minutes with Claude, versus 45 minutes of direct reading to extract the main points.
Audiovisual creation: Runway, CapCut AI, Remove.bg
The creative sector is where AI creates the most opportunity — and the most disruption. What professional artists spent years learning, an amateur can now simulate in a few clicks. It's fascinating, unsettling, and you'd do well to keep both eyes open.
Runway ML — Accessible generative video
Price: Free (limited credits) / Standard at $12/month / Pro at $28/month.
Runway Gen-2 and Gen-3 Alpha allow you to generate video clips of a few seconds from a text prompt or reference image. Quality in late 2024 is impressive for illustrative uses: atmospheres, visual concepts, stylised transitions. It's far from cinema — movement inconsistencies on people are still frequent — but for social media content or visual presentation elements, the quality-to-effort ratio is unbeatable.
I use Runway primarily for two things: generating animated backgrounds for YouTube videos, and producing illustrative clips for articles. The result looks professional for someone with no motion design training.
CapCut AI — AI-assisted video editing for content creators
Price: Free (with limitations and watermark) / Pro at $9.99/month.
CapCut (owned by ByteDance, TikTok's parent company) has integrated massive AI features into its video editing tool: automatic silence removal, subtitle generation with karaoke display, background replacement, automatic style effects, and a voice cloning function that genuinely deserves careful ethical reflection.
For content creators on TikTok, Instagram Reels, or YouTube Shorts, CapCut has become indispensable. Productivity on short edits is multiplied 3–5x according to regular users. Automatic silence removal alone saves 20–30% of editing time.
An important note on data: CapCut is a ByteDance application. If you work with sensitive or professional content, think about the data you're sending to their servers. The ICO (Information Commissioner's Office) has issued guidance on data transfers to third countries outside the UK under UK GDPR — it's worth checking before uploading client-facing content.
Remove.bg — The perfect AI single-use case
Price: Free (low resolution) / from £7/month for commercial high-resolution use.
Remove.bg does one thing: remove the background from an image. And it does it perfectly, in 5 seconds, without training or professional software. Before this tool, this operation required Photoshop, the "Select Subject" tool magic, and still 5–10 minutes of manual edge retouching for complex subjects (hair, fur, transparency).
This is the perfect example of specialised AI beating generalist AI: Remove.bg does one thing, perfectly, instantly. You don't need ChatGPT or Midjourney for that. These highly targeted tools are often the most efficient in daily life.
Before subscribing to a general AI suite at £16–25/month, list the 5 repetitive tasks that waste the most of your time. There's probably a specialised tool that solves them precisely — often free or at £5/month — and will be 10 times more effective than a poorly configured generalist tool. Remove.bg, Otter.ai, Reclaim for scheduling slots — this trio can transform a working week for under £20/month combined.
AI in everyday life: voice assistants and the smart home
This is the domain where the technological promise has diverged most from everyday reality — until very recently.
The evolution of voice assistants: finally useful?
Siri, Google Assistant, Alexa: we've all used them to set timers and turn off lights. The historical problem was their incomprehension of complex or unscripted natural language requests. In 2024, that's changing.
Google Assistant with Gemini integration: Google has progressively integrated Gemini into its assistant. On Android, this translates into markedly improved understanding of contextual requests and the ability to reason about multi-step tasks. You can say "send a message to my sister to say I'll be 30 minutes late" and the assistant understands, proposes the message, and waits for your confirmation. This was impossible reliably three years ago.
Apple Intelligence (iOS 18, late 2024): Apple launched its on-device AI infrastructure with iOS 18, integrated into Siri and native apps. Early feedback is mixed — the most advanced features (deep personal context understanding, cross-app actions) are still being progressively rolled out in the UK. Apple's argument: on-device processing preserves privacy. That's a real and distinctive advantage.
Amazon Alexa: Amazon announced a major Alexa overhaul with LLM capabilities in 2023. In practice, deployment has been uneven. Alexa remains excellent for home automation commands and smart home integrations, but is losing ground on natural conversation compared to Google Gemini.
Smart home and AI: what's actually worth it
AI in the smart home concretely translates to:
- Intelligent automations: routines that adapt to your actual behaviour rather than following a fixed schedule. The Hive or Netatmo thermostat learns your heating habits and optimises automatically. Average saving: 20–30% on heating bills according to manufacturers (figures to take with caution — conditions vary enormously by property type).
- Predictive security: modern security cameras (Arlo, Nest, Ring) distinguish a human, an animal, a car — not just motion. Result: fewer false alerts, more relevant notifications.
- Anticipating needs: Samsung Family Hub fridges can track their contents and suggest recipes based on what's inside. Theoretically brilliant. In practice, the AI features of these appliances often go underused — the setup is tedious and results variable.
My overall view on AI smart home: basic automations (lighting, heating, scheduled plugs) deliver real, measurable value. More complex AI features often require more configuration than they save in time during the first few months. Start simple, add complexity once the basics work reliably.
Honest limitations: hallucinations, ethics, environment, privacy
I'd be doing you a disservice if I only wrote about where AI works. Here are the real limitations that every honest user needs to know.
Hallucinations: an unresolved reality
Language models "hallucinate" — they generate false information with the same confidence as true information. This isn't a bug that will be fixed in the next version; it's an architectural characteristic of current LLMs. The model predicts the next probable token; it doesn't "know" whether information is true.
In practice, this means:
- Never directly use a book or article citation generated by AI without verification — invented references are common.
- Never trust precise numerical data (statistics, dates, prices) without verifying from a primary source.
- For high-stakes domains (medicine, law, finance), systematically treat AI content as a first draft to be validated by a human expert.
The Ada Lovelace Institute's 2024 report on AI systems and information quality explicitly recommended that organisations handling public information establish systematic verification protocols for all AI-generated content. That's common sense that many users ignore at their peril.
The ethical question of creative displacement
I'm not going to pretend this isn't a problem. Illustrators, stock photographers, web copywriters, voice actors — all these professionals are seeing their livelihoods threatened by tools trained on their own work, often without consent or compensation.
The Getty Images lawsuit against Stability AI (filed in the UK High Court in early 2023), the ongoing class actions in the US against OpenAI and others — the legal landscape is genuinely unsettled. The UK's proposed approach to AI and copyright, currently under consultation, may diverge from the EU AI Act, creating additional complexity for creators and businesses operating across both jurisdictions.
My personal position: I use these tools because they're useful, while being aware that I'm benefiting from a system that hasn't yet resolved its fundamental ethical questions. Actively supporting human artists — buying their original work, citing them, recommending them — is a partial way to offset the impact. It's not enough, but it's something.
The environmental cost — let's talk about this honestly
Generating an image with Midjourney consumes approximately 10 times more energy than a Google search. Training GPT-4 consumed the equivalent of 500 tonnes of CO2 according to OpenAI's own estimates (likely an underestimate according to independent analyses). The IEA (International Energy Agency) predicts that data centre electricity consumption for AI will triple between 2023 and 2026.
This isn't a reason not to use these tools. It is a reason to use them intentionally rather than compulsively. Generating 50 variations of an image to choose one is not the same as generating 2 variations. Every request has a cost that doesn't appear on your bill but exists in the physical world.
Privacy and personal data
Every request you send to ChatGPT, Claude, or Perplexity is sent to servers — often in the United States. By default, OpenAI was using conversations to improve its models (disableable in settings). Under UK GDPR, you have rights over how your data is used — but exercising those rights requires knowing where to look in the settings and understanding what you agreed to in the terms of service.
Practical rules:
- Never paste personal data (names, email addresses, phone numbers, medical information) into prompts destined for cloud services.
- For professionals handling client data, verify the UK GDPR compliance of the service being used — or use on-premise solutions.
- Disable conversation history in settings if privacy matters to you. Both ChatGPT and Claude offer this option.
- Explore open-source local alternatives (Ollama + Llama for advanced users) for fully local use with no data transfer.
- The ICO's guidance on AI and data protection (updated 2024) is worth reading if you're using AI tools in a professional context with UK clients or employees.
How to choose your tools without drowning
The AI tools market is exploding: several thousand new services launched in 2023–2024. Most won't survive to 2026. How do you avoid being overwhelmed?
The decision framework: 5 questions
- What precise task does it solve? If you can't answer in one sentence, the tool might be a gadget.
- How long does it take to configure? If setup takes longer than 6 months of estimated time savings, question whether it's worth it.
- Who has access to your data? Read the terms of service, genuinely, at least for tools you'll use with sensitive professional or personal content.
- Is there a usable free version? The market is so competitive that most serious tools offer generous free tiers. Test before paying.
- What happens if the service disappears? With hundreds of AI startups founded since 2022, some will close. Choose tools whose data is exportable in standard formats.
My personal stack — what I actually use
Here's what I genuinely use day-to-day, without marketing fluff:
- Claude Pro (£14/month): long-form writing, document analysis, structured brainstorming.
- Perplexity free: daily synthetic web research. The Pro version hasn't seemed necessary for my usage.
- Otter.ai Pro ($10/month): transcription of all my client meetings.
- Reclaim AI free: automatic scheduling of deep work slots.
- Remove.bg: on demand, using the small free credit allocation.
- DALL-E 3 via ChatGPT Plus: I have ChatGPT Plus for projects requiring GPT-4o + image generation. I use it alongside Claude, not instead of it.
Total: approximately £40/month. That's not nothing. In return, I estimate I save 4–6 hours of work per week — which more than justifies the investment at my hourly rate. But that calculation is very personal; do the maths for your own situation.
Tools I tested and abandoned
For balance:
- Jasper AI: expensive ($49/month) for results that Claude achieves better at £14.
- Copy.ai: useful for short marketing copy, but redundant with ChatGPT Plus.
- Synthesia: videos with AI avatars — interesting conceptually, but the "uncanny valley" quality of avatars undermines authenticity in most contexts.
- Motion: too complex to maintain for my workflow. Reclaim is simpler and does 80% of the work.
Do you need to pay for a genuinely useful AI tool?
No, not necessarily. The free version of ChatGPT (GPT-3.5 and now GPT-4o with limits), the free version of Claude, and the free version of Perplexity cover the majority of everyday needs. Start free, identify the limitations that genuinely frustrate you, and only pay to remove those specific friction points. Subscriptions at £16–20/month only justify themselves if you're using the tool intensively — several hours a week, every week.
Can AI replace a professional copywriter?
For standardised, low-creative-value content (basic product descriptions, generic location summaries, document summaries), LLMs do replace some of the work of junior writers. For journalism, creative writing with a distinctive voice, expert analysis, and anything requiring investigation or quality judgement — no. AI is an accelerator for competent human authors, not a replacement for them. Articles produced entirely by AI without human oversight are recognisable in their genericity, and Google's algorithms are increasingly deprioritising them. Which? and The Guardian Tech have both published investigations into AI-generated content farms — the results are not flattering.
Can AI do my tax return or bookkeeping?
With important nuances. Claude or ChatGPT can help you understand tax concepts, structure your bookkeeping, or draft queries to your accountant. They cannot replace a qualified accountant for official filings — and you shouldn't trust a LLM that can hallucinate amounts or tax rules with your HMRC submissions. Dedicated tools like FreeAgent, QuickBooks, or Xero integrate AI within a controlled legal and accounting framework — that's the right approach for professional financial management. For self-assessment help, HMRC's own guidance and a qualified accountant remain the authoritative sources.
Is my data safe with ChatGPT or Claude?
The honest answer: it depends what you mean by "safe". Data isn't "stolen" or publicly exposed. However, it transits through servers of US companies subject to US law (including government access provisions), may be used to improve models if you haven't disabled that option, and is accessible to company teams under conditions specified in their terms of service. Under UK GDPR, you have rights over your data — but exercising them requires knowing where to look. For sensitive personal information (health, financial data, client information), don't use these cloud tools. Disable conversation history in settings. Consider open-source local alternatives for the most sensitive uses.
How do you avoid becoming dependent on AI tools?
It's a legitimate question. Cognitive dependence develops when you delegate tasks to AI without understanding what's being produced — which progressively weakens the corresponding skills. For writing, I continue to draft manually first and then use AI to improve, rather than the reverse. For research, I continue to read primary sources rather than only accepting syntheses. Use AI to amplify your skills, not replace them. The Which? Tech survey of 2024 found that users who reported the highest satisfaction with AI tools were those who used them as "skill enhancers" rather than "skill replacements" — a distinction worth keeping in mind.
- ICO — AI and Data Protection Guidance — UK Information Commissioner's Office guidance on AI systems and UK GDPR compliance
- Ada Lovelace Institute — Foundation Models: Opportunities, Risks and Governance — Independent analysis of large language model capabilities, limitations and governance gaps
- IEA — Electricity 2024 Report — International Energy Agency projections on AI data centre electricity consumption
- The Guardian — AI Technology — Ongoing investigative journalism on AI tools, copyright, and impact on creative industries
- Which? — AI Assistants — Independent consumer testing and user satisfaction surveys on AI tools in the UK
- EU AI Act — Official Documentation — European regulatory framework on artificial intelligence, including obligations relevant to UK-adjacent markets