My stack, with receipts

I am going to do something that most "my AI stack" articles avoid: tell you exactly what I pay, exactly what I use each tool for, and exactly why I have not cancelled any of them. No vague "it makes me more productive" hand-waving. Specific use cases, specific numbers.

Total monthly spend: about forty-five pounds. Here is the breakdown.

Claude Pro - sixteen pounds per month

claude is my primary writing tool and it is worth every penny. I use it for:

First drafts: I write about 15 articles per month for various clients and publications. Claude helps with first drafts on about half of them. Not by writing the article for me, but by helping me structure my thoughts, suggesting angles I had not considered, and getting words on the page when I am staring at a blank screen.

My workflow is specific: I give Claude my outline, key points, and the tone I want. It gives me a rough draft. I then rewrite about 70% of it, keeping the structure and a few particularly well-phrased sentences. The output is mine, but the process is faster.

Editing pass: I paste my finished drafts into Claude and ask it to identify weak arguments, unclear sentences, and logical gaps. This is different from Grammarly (which handles grammar and style). Claude catches the big-picture issues. "This paragraph contradicts what you said in section two" or "your evidence for this claim is thin." That kind of feedback used to require a human editor, and while Claude is not as good as my best editor, it is available at midnight on a Sunday when I am finishing a deadline piece.

Research synthesis: When I am working with multiple sources, I paste them into Claude and ask it to identify agreements, contradictions, and gaps. The 200K context window means I can feed it ten articles on a topic and get a coherent synthesis.

Why I have not cancelled: Claude's writing quality is noticeably better than ChatGPT's for my use case. The prose is more natural, the suggestions are more nuanced, and the editing feedback is more perceptive. I tried switching to ChatGPT Plus for a month and the quality difference was obvious enough that I switched back.

Grammarly Premium - twelve pounds per month

grammarly is the most boring tool in my stack and probably the most valuable.

I used to think I did not need a grammar checker. I have a degree in English. I have been writing professionally for eight years. My grammar is fine. Then I installed Grammarly and discovered that I consistently misuse "which" and "that," my sentences average 28 words (way too long), and I overuse the word "actually" to a degree that would embarrass me if I thought about it (I used it three times in this article's first draft before Grammarly flagged it).

Grammarly runs in the background across everything I write. Browser extension, Google Docs plugin, desktop app. It catches errors in emails, Slack messages, article drafts, and social media posts. The tone detection is particularly useful for client emails where I need to sound professional without being cold.

Why I have not cancelled: The errors Grammarly catches are the kind that make you look careless. A misplaced comma in an article is mildly embarrassing. A typo in a client proposal can cost you the job. Twelve pounds per month for that insurance is absurdly cheap.

Perplexity Pro - sixteen pounds per month

perplexity replaced Google for about 80% of my research. The difference is speed and citation quality.

When I am researching an article, I need facts I can verify. Perplexity gives me direct answers with numbered citations that link to the original sources. Google gives me ten blue links and makes me do the synthesis myself. For the kind of fact-heavy writing I do, Perplexity saves me at least an hour per article in research time.

The Pro tier is worth it because the standard model sometimes gives shallow answers. The Pro search uses more powerful models and produces more thorough, nuanced responses. For the kind of technical and industry research I do, that depth matters.

Why I have not cancelled: At fifteen articles per month, saving an hour each on research is fifteen hours. At my hourly rate, that is worth considerably more than sixteen pounds. The ROI is obvious.

Notion AI - seven pounds per month (AI add-on)

notion-ai is the tool I am least sure about keeping. I use Notion as my project management and notes platform (the base plan is free), and the AI add-on adds features like summarising meeting notes, generating action items from long documents, and helping with content organisation.

The honest truth is that I could probably do everything Notion AI does by copying text into Claude. The convenience of having it embedded in my notes app is real but hard to quantify. It saves me maybe 30 minutes per week, which is real time but not dramatic.

Why I have not cancelled yet: Convenience. Having AI in my notes app means I actually use it for quick tasks that I would not bother opening Claude for. Summarising a rambling meeting transcript, extracting key dates from a long email thread, generating a quick project brief from my scattered notes. These are small tasks that add up.

Why I might cancel: If Claude ever adds proper integration with note-taking apps, Notion AI becomes redundant. I am watching that space.

What I have tried and cancelled

Jasper ($49/month): Excellent for marketing copy but I do not write enough marketing copy to justify the price. If you are a full-time marketing writer, it is worth considering.

Copy.ai ($49/month): Good for short-form content but the output quality for long-form is not high enough. Everything it generates needs so much editing that the time savings are minimal.

Writesonic ($16/month): Cheaper than Jasper and Copy.ai but noticeably lower quality. Fine for SEO content mills, not for anything where quality matters.

Hemingway Editor ($10/month): I loved the concept but Grammarly covers the same ground plus much more. Could not justify paying for both.

The total picture

Monthly spend: forty-five pounds Monthly output: 15+ articles, 50+ client emails, various other writing tasks Estimated time saved: 25-30 hours per month Effective cost per hour saved: less than two pounds

That is the calculation that matters. Not "is this tool impressive?" but "does this tool save me enough time to justify its cost?" For my four tools, the answer is clearly yes. Your stack will probably look different because your work is different. But the framework for evaluating is the same: time saved times your hourly rate should comfortably exceed the subscription cost. If it does not, cancel it.