Every list of best AI tools for content is the same list. Copied feature tables, prices lifted off the pricing page, and nowhere a person who used the thing for six months. A model gains nothing from that, because it can already produce it. What I can add is the column nobody fills in: what broke. These are the nine tools I run articles, images, video and publishing on, and if you have to pick one it’s Claude Code, because it is the only one that sees my whole project instead of a single task.
The nine side by side
| Tool | What I put it to work on | Price | What went wrong | Where this knowledge comes from |
|---|---|---|---|---|
| Claude Code | Writing and publishing inside my sites’ repos, several files at once | Included in Claude Pro, 17 dollars a month annually or 20 monthly | The branch collision between parallel sessions, set out in the piece on Code, Cowork and Dispatch. Every session now runs in its own worktree | 32 published pieces |
| Claude Cowork | Recurring tasks with no repo, running on with my laptop shut | Same subscription, heavier limits from Max at 100 dollars a month | You hit the five-hour ceiling on exactly the day things are flowing | 6 published pieces |
| Bonsai Image, local | Every cover on this site, in one consistent style | Free and open source, costs disk and compute | The model invented readable letters on the image and turned them into a fake brand name resembling mine; a half-installed environment that started some days and not others | 3 published pieces |
| Blotato | Publishing to every social channel at once, plus DM automations | Subscription already in place | A post reporting “published” may never have existed, and a post reporting “failed” may well have gone live; above 200 scheduled posts the plan refuses with an explicit error | Project notes, no blog piece |
| ManyChat | Comment-to-DM funnels on Instagram | PRO subscription, frozen since August 2026 | Of 207 flows, 58 ever fired, 154 times in total; 139 live flows with a keyword sit at zero. That is not a keyword problem but a reach problem | Project notes, no blog piece |
| Remotion | Rendering video and reels from code | Free for my use, open source | 4K source material sometimes renders entirely black with no error, exit code 0 and a one-and-a-half megabyte file; the yuvj420p colour profile produces video QuickTime refuses to open | Project notes, no blog piece |
| clipify plus Whisper | Cutting long video into vertical clips with captions | Free and local | The ffmpeg install lacks libass, so burning subtitles fails; I now render them as a transparent PNG layer via PIL and overlay that | Project notes, no blog piece |
| ChatGPT, Perplexity, Gemini | Loose questions and research alongside the rest | Free up to a few tens of euros a month, depending on plan | Nothing recorded | Used, but never written about |
| Descript | Editing audio and video | Subscription | Nothing recorded | Used, but never written about |
That last column is why this list is worth anything. For seven tools I can say what went wrong and how I worked around it. For two I can’t, so I say that instead of inventing a verdict.
What to weigh when choosing
Does it see your project, or only your question? This is the sharpest distinction. Claude Code reads the folder and adopts the conventions already there. A chat window starts from zero every time. For loose tasks that changes nothing, for a site with hundreds of files it changes everything.
What happens when it fails, and do you notice? This is where these tools really differ. Remotion renders black with exit 0. Blotato reports “published” for a post that isn’t there. Both fail silently, and silent failures cost more time than loud ones.
Does it run locally or on someone else’s machine? Bonsai, Remotion, clipify and Whisper run on mine. Claude Code, Cowork, Blotato and ManyChat do not. For covers that hardly matters, for material that isn’t public yet it does.
What does leaving cost? With a local model, nothing: it sits on your disk and keeps working. With a service you lose the workflow built around it. ManyChat is the example here: frozen, but the subscription runs on as long as live flows sit inside it.
The tools I learned most from
Blotato is the most used and the hardest to trust. The status message is not reality. I have seen posts labelled “published” that existed nowhere, and posts labelled “failed” that went live regardless. Anyone acting on the label publishes twice or not at all. The habit that stuck: always find the post on the platform itself before you fix anything.
ManyChat produced the most uncomfortable number on this list. More than two thirds of the flows never did anything. That is not proof that comment-to-DM doesn’t work, but it is proof that my reach is the bottleneck rather than the keyword. Since August I build no new flows.
Bonsai taught me most about how image models work, precisely because it went wrong. Why I do that locally and why I keep at it is set out in two earlier pieces. The model wrote letters onto images where I asked for none, and turned them into a brand name resembling mine. Since then I write prompts that name no readable surface: no label, no sign, no text.
The caveat: these are nine dependencies
Four of these nine run locally, five do not. Where that line actually pays off I set out earlier in local model or cloud. Those five together decide whether I can publish at all, and over none of the five do I have any say about price, limits or continued existence. The Blotato cap of 200 scheduled posts is a small example: not a bug, just a plan boundary that suddenly becomes the floor in a busy month.
There is an energy side too that I rarely see named. Rendering locally costs my battery and my time; rendering in the cloud costs someone else’s power, and that disappears from view the moment it isn’t on your own bill.
And the point that weighs heaviest: the more of this chain runs automatically, the less often I actually read something before it goes out. Blotato politely asks for approval. That approval becomes a click once you give it often enough. The brake is in the tool, the habit is on me.
Frequently asked questions
Which AI tool is best for content production?
For work that lives in a project folder, Claude Code, because it reads the structure instead of starting over each time. For publishing across channels, Blotato is the one I use daily, with the caveat that you should not trust its status messages.
Can you make images for your site locally without a subscription?
Yes. Bonsai Image runs on a laptop and is open source. Budget several gigabytes of disk and an evening of setup before the pipeline stands.
Do comment-to-DM funnels work on Instagram?
Barely, in my case. Of 207 flows, 58 ever fired, 154 times in total, and 139 live flows with a keyword sit at zero. The bottleneck turned out to be reach, not the keyword you pick.
Why is there no price comparison?
Because the prices that matter sit in one subscription for me and the rest runs free and locally. A table of monthly fees would suggest the choice is about money, while for this stack it is about reliability and about ownership.
Figures on ManyChat and Blotato come from my own project log, last updated in August 2026. Anthropic prices were checked on the pricing page on 9 August 2026 and change regularly.