NotebookLM Data Tables: I Fed It a Messy Spreadsheet
NotebookLM Data Tables promises to turn messy data into clean tables. I tried it with a real messy spreadsheet. The catch is at the input, not the AI.
Create
July 7, 2026 · 5 min read
The short version: I ran three real writing tasks through Claude myself and checked the five other most-recommended AI writing tools against verified 2026 pricing pages and the loudest complaints on G2, Reddit, and Trustpilot, instead of pretending to have six paid subscriptions open at once. Claude produced the cleanest first-draft prose. Jasper wins if you need one locked brand voice across a team. Sudowrite is worth it only for fiction, and Copy.ai just got a lot more expensive for solo writers than its reputation suggests.
This is for anyone picking a paid AI writing tool for blog posts, marketing copy, or a novel. Let's be honest: instead of faking a "blind" test across six accounts I don't own, I ran the real test through the one tool I have and checked the rest against what independent reviewers already measured.
I wrote three short pieces with Claude myself: a 50-word product description, a tone rewrite of a stiff corporate paragraph, and a two-sentence blog hook from a literal prompt. The other five tools I checked against verified pricing pages and independent reviews.
Claude's first attempt was bad in a way I recognized immediately: "an innovative reusable bottle that brings a premium, must-have upgrade to your daily routine... state-of-the-art insulation meets effortless convenience."
Zero facts, all filler.
The second attempt, after I banned marketing language in the prompt, read like a person wrote it: "HydroLoop keeps drinks cold for 24 hours and hot for 12, in a stainless shell that fits your bag's side pocket. The flip-lid opens with one thumb, so you can drink without stopping mid-run. Fill it in the morning. Forget about it until dinner."
The tone rewrite worked on the first try. I fed Claude a stiff, jargon-heavy paragraph about "utilizing innovative solutions" to "facilitate enhanced stakeholder engagement," and it came back with: "We want people to enjoy talking to us, whether that's over email, on a support call, or in a five-star review. So we're finding better ways to listen and respond at every point where someone interacts with our brand." No filler, and it cut the word count almost in half.
Write a 2-sentence opening hook for a blog post titled "How to start a podcast in a weekend." First sentence: a concrete, slightly odd detail about where I recorded my first episode. Second sentence: what the reader gets by the end, stated plainly. No throat-clearing, no rhetorical question.
The output: "I recorded my first podcast episode inside a closet full of winter coats, because it was the quietest room in my apartment. By the end of this post you'll have the same six-step setup I used, from a $40 microphone to a live episode on Spotify, in one weekend."
Claude is Anthropic's general-purpose assistant, the only tool here I tested myself. Pricing as of July 2026: free tier, Pro at $20 a month ($17 annually), Max plans from $100. Two of my three tasks came back clean; one needed a redo after a cliché-heavy draft. An independent eight-platform comparison from eesel AI reached the same conclusion: Claude's prose reads like a person wrote it. It's for solo writers who want a clean draft with no training period, not a team that needs five writers locked to one brand voice. Verdict: 9/10.
ChatGPT is OpenAI's chatbot, still the most-used AI tool worldwide. Pricing as of July 2026: free, Go at $8 a month, Plus at $20, Pro at $200. I didn't test it here; I compared it against Claude and Gemini on ten real tasks in ChatGPT vs Claude vs Gemini: I tested all three on real work. The gap shows up there too: 2026 analyses flag the em dash as a persistent AI signature in its default output, the same mark our house style bans outright. It's for the broadest feature set in one app, not the fastest route to prose that doesn't read like AI wrote it. Verdict: 7/10.
Jasper trains one consistent brand voice across a team; it's not a general chatbot. Pricing as of July 2026: $59 a month annually ($69 monthly) for one seat on Pro, custom Business tier for more. G2 reviewers give Jasper 4.7 out of 5 across 1,200-plus reviews, with brand-voice consistency the most-praised feature. Those same reviewers call output "cookie-cutter" until that voice is trained, which takes time you won't have on day one. It's for a team writing at volume, not someone who wants a good draft this afternoon. Verdict: 6/10.
Copy.ai built its early reputation as a cheap, all-in-one writer for solo marketers. That's changed: its pricing page now shows no permanent free plan, just a "Chat" tier at $24 a month annually ($29 monthly) for five seats, with the next tier jumping straight to $1,000 a month. Trustpilot, which bars incentivized ratings, puts Copy.ai at 1.9 out of 5; Capterra reviewers report needing 40 to 60 percent rewriting on longer pieces. One detail stood out: reviews quote its default output opening with "In the fast-paced world of digital marketing," the exact kind of line our banned list would reject. It's for a go-to-market team with a real budget, not a freelancer. Verdict: 5/10.
Sudowrite is built for fiction, with a model called Muse trained on creative prose, not marketing copy. Pricing as of July 2026 starts at $10 a month annually ($19 monthly) for 225,000 credits, up to $44 annually ($59 monthly) for 2,000,000 rollover credits on Max. Its "Show Not Tell" rewrite is the most praised feature in reviewer feedback, useful for fixing the exposition-heavy paragraphs every first draft has. Reviewers also warn that leaning on it too hard can flatten your own voice. It's for novelists who want a second brain for scene-level prose, not for blog posts or ad copy. Verdict: 8/10, fiction only.
Grammarly checks and rewrites text you've already written; it doesn't draft an article from a blank page. The free tier includes 100 AI prompts a month, and Pro costs $12 a month annually ($30 monthly) with 2,000 prompts per member. Grammarly's own support docs put it plainly: it's not your content factory, it's your content editor. That matches what I'd tell you. It's for the last pass on something already drafted, not for starting from nothing. Verdict: 7/10.
| Tool | Starting price | Best for | One-day score |
|---|---|---|---|
| Claude | $20/mo | Solo writers, clean first drafts | 9/10 |
| ChatGPT | $20/mo | Broadest feature set, one app | 7/10 |
| Jasper | $59/mo | Team brand voice at scale | 6/10 |
| Copy.ai | $24/mo | GTM teams with a real budget | 5/10 |
| Sudowrite | $10/mo | Fiction writers only | 8/10 |
| Grammarly | $12/mo | Editing an existing draft | 7/10 |
Pick based on what you're starting from, not the brand you've heard of most. Writing solo for a clean draft today: start with Claude or ChatGPT. Scaling one brand voice across five writers: Jasper's training time pays off. Deep into a novel: Sudowrite's fiction-only focus wins. Already juggling more than one AI subscription: my running breakdown of what AI actually costs per month is worth reading first. Just polishing text you already wrote: skip the drafting tool and get Grammarly.
Would I pay for any of these myself? Claude, without hesitation. The other five earn their price only in a specific scenario, and Copy.ai's new pricing means it no longer earns the "cheap and cheerful" reputation it still gets credit for online.
The tool that saved me time this week wasn't the flashiest one. It needed the least cleanup.
NotebookLM Data Tables promises to turn messy data into clean tables. I tried it with a real messy spreadsheet. The catch is at the input, not the AI.
I gave five AI writing tools one brief: write my bio. Each broke differently. Where ChatGPT, Jasper, Copy.ai, Grammarly and Sudowrite fall short.
A Sora 2 test review that skips the hype: a verified map of where its physics promise breaks (hands, flips, text) and why the app closed in 2026.