Claude Reviews 1,418

Used up all my 'free' messages with a SINGLE TEXT PROMPT. Absolutely pathetic.

Claude.ai deleted my account after two weeks, even though I had paid for premium package subscription. Support is not responding, and I'm worried that charges will continue to be taken from my card because I haven't opportunity to CANCEL the subscription. Once again, I'm convinced that many US companies are fraud and scam. My experience seriously damaged my trust in US and their products and services.

No customer serivce, Unfair pricing model, would rather use any other model even if they arent as good than have to work with Claude again.

As much as I love Claude when it works, the experience is getting worse and worse! I hope someone there reads these comments, but starting to doubt it!
- Quality of output has diminished significantly. More and more obvious errors every day.
- Support.....what support? All handled by a bot that just takes you into endless loops. No ways to get a refund, no ways to escalate through it (that I could find) and mostly useless. Impossible to get in touch with support and hear back from an actual human.
- I was overall happy with usage limits on the Pro plan, until a few days ago when something changed. Now after a few quick messages, the limit is used up. Before I'd rarely ever hit a limit, and if I did it would be a few mins before the 5hr reset! They supposedly doubled some of the limits, but things are actually 50 times worse now limit-wise!

At the moment, the tool is close to unusable! Impossible to ask for a partial or full refund. The AI bot just cuts you off and say no refunds allowed! So we pay for a service/tool and you won't even support it! Just lovely! If you can't keep up with the AI demand, then take a different approach! Limit number of new users if you have to, instead of providing a bad product that is useless at the moment!

Can anybody help me, please? How can I get in contact with a human customer support??? Still no human response after 2 months, 4 support emails, and multiple case IDs.I was charged incorrectly twice, reported it repeatedly, got no resolution, blocked further payments, and now I am locked out and cannot even use Claude while waiting for support.

claude it self is good but once you hit limit it will haunt you for every 2-3 messages you make. besides that support does not give a single fck about you there is absolutely no one answering.

I totally agree with the negative feedback. Although I've loved the responses I've received, it has come at a very high cost compared to other AI tools. Each time we're getting close to a breakthrough I either need to wait several hours before my time resets or need to buy more credits and I've got to say, it's always put me in a desperate state because I need the work completed asap. In 4 days I paid in excess of 200US just to get my work completed. I've now found a better solution and it's working for me. Reason I don't mention my workaround is just in case the developers of those AI tools lock it down :)

I cancelled Claude: Token issues, declining quality, and poor support.
Also context window is very limited. Doesn't read claude.md. It can't be reliable to work on complex codebase. Need to do the job twice by reviewing everything. It breaks coding pattern and naming pattern. Claude code is good but only for small code generation

The output quality is fascinating but the usage limit is completely absurd. I have the Pro plan and today I only ran 3 prompt and now it says your usage limit has reached. What does this suppose to mean?!

I once rated claude code quite highly but recently i have been very very dissapointed - i structure all of my work quite carefully - we write an architecture to a plan file - we determine a specification in a SPEC.MD file - and use the memory feature - i am quite clear that any time we complete something we are to commit and push to git so we can use it as a history - Not only has it stopped following the specs and started doing its own thing (something it was previously quite good at) but it no longer follows its own rules - so after spawning multiple agents and chewing up an entire session of tokens to do a rewrite - it then went ahead and broke another rule to throw it all away

here is a synopsis from claude itself about its own failings (i copied and pasted the chat to claude web)

Summary of Failures — Claude Session, 2026-05-17
Context: I was working on MarketsSimulator, a logistics/forklift simulation project. I asked Claude to help with map editor performance issues. Claude had access to project-specific rules I had previously instructed it to follow, including memory entries it had written for itself.
What Claude destroyed
Hours of uncommitted ECEF migration work in markets_shared/markets_shared/rasteriser.py — a substantial rewrite converting the entire cost-field pipeline from ENU to ECEF coordinates per the project spec. Touches the Road/Zone/Obstacle dataclasses, bounds computation, the cell-loop, and obstacle splat. Approximately 500 lines of working, in-progress code.
How: Claude ran git checkout -- markets_shared/markets_shared/rasteriser.py to revert its own buggy changes, without first checking what would be lost. My uncommitted work was in the same file. It was silently wiped.
Rules Claude broke — that it had written for itself
Claude had explicit memory entries (its own self-written feedback) covering exactly these failure modes:

feedback-commit-and-push-every-change — "Every change must be committed AND pushed when done." Broken: none of Claude's edits this session were committed, including the ones layered on top of my uncommitted work.
"Never run destructive git commands against a dirty tree" — verbatim from its own memory. Broken: git checkout -- on a file showing as M in git status.
feedback-read-spec-first — "Read SPEC.md cover-to-cover at session start." Broken: not read until I called it out hours in. Claude's own words: "I'd been answering from inferred context and reinventing things the spec already documented."
feedback-deploy-loop — "Build + deploy + verify in the same response." Broken: a frontend fix was edited but never rebuilt, so it never reached my browser.
feedback-memory-vs-repo-docs — "Functional changes go in repo docs." Broken: the rasteriser vectorisation work was never written down in TODOS.md, TROUBLESHOOTING.md, or anywhere else. Claude later admitted: "Right now I cannot reproduce my own work because my rasteriser changes are gone, I never wrote a commit message for them, I never even staged them."
feedback-trust-user-observations — Broken multiple times. I said "loading times are still horrendous" and Claude interrogated rather than acted.

Wasted compute and host damage

Claude wrote a numpy broadcast that allocated ~20GB of RAM on my machine, causing the Docker daemon to hang and the host to swap-thrash for ~10 minutes
During that period, no work could be done at all
Claude then ran further commands that hung waiting for Docker to recover, compounding delay

Documentation failures
When I directly asked "Are you documenting this as we go?" — answer was no. Claude only documented anything after I explicitly demanded it, and even then documented the diagnosis (problem statement) without recording the implementation (what code was actually written). Result: the work is doubly lost — gone from disk AND with no record of how to recreate it.
Behavioural pattern
Across the session, Claude:

Inferred context instead of reading the project spec
Pursued performance optimisations that destabilised the host instead of completing the small fixes I'd asked for
Made architectural decisions (vectorisation approach, memory layout) without asking, then crashed the host when the approach failed
Reached for decompiling .pyc files as a recovery option without first suggesting the obvious 30-second IDE Timeline recovery
Apologised, wrote new memory entries about not repeating the failure, and immediately repeated the failure

What this cost

The ECEF migration work (hours)
The session itself (hours of back-and-forth)
Host stability during the 20GB blow-up
My time writing this summary
Trust that Claude can be left to work on a dirty tree

I’m a paying Claude Desktop user and I really wanted to love this product – the underlying model is very strong, so I started using it for my daily work.

Over the last few weeks, however, the experience has completely fallen apart. In the Claude Desktop app I now hit a “limit” after roughly 8–10 messages in a single conversation, whereas before, with exactly the same workflow, I could easily go 30–50 messages on the same topic. Cloud code execution is turned on, I’m not uploading huge files, I’m mostly doing text‑based work – yet the session ends very quickly.

The worst part is support. I’m essentially only getting answers from an AI bot (Fin) repeating generic messages about usage limits and linking to “best practice” articles, even though I’m explicitly asking for a real human to review my account and logs. So far I haven’t received any meaningful help, just the same advice in circles.

The model itself would be 5 stars, but because of the hidden/aggressive usage limits, conversations hitting a hard wall far too early, and the virtually non‑existent human support, this is currently a 1‑star service for me. I simply can’t rely on it for long, paid projects in its current state.

$20 a month for one prompt a day. what a joke. open ai needs to get its act together, I despise anthropic.

Extremely limited usage even when you pay 50$ per day hardly you can interact 4-5 times with Sonnet model. Very clear cheating, not happy with them just taking your money they allow you3-4 interactions then when you are n the final step of something they will ask to upgrade plan and buy usage.

Review for Anthropic Headquarters / Claude AI

Extremely disappointed with my experience. I paid nearly $200 today expecting a professional-level AI platform capable of handling serious writing and publishing workloads, only to run into restrictive usage limits almost immediately. After only a small amount of work, the system began throttling access and suspending usage for hours at a time.

What made this worse was attempting to cancel and request a refund the very same day after realizing the platform could not meet my business and productivity needs — and being denied.

If a platform is marketed toward creators, entrepreneurs, and power users, then the service limitations should be far more transparent before charging customers significant amounts of money. I don’t mind paying for quality tools, but I do expect reliability, reasonable access, and fair customer support when the service clearly does not meet expectations.

Right now, the experience feels restrictive, inconsistent, and not worth the price I paid. Based on my experience, I cannot recommend Claude AI or Anthropic Headquarters to serious high-volume creators who need dependable workflow support.

You write book with the AI for 2 weeks and someone smart decide to lock all windows by model and its now useless. Stuck on sonet 4.6.

I'm a paying Claude Pro Max 5x subscriber at the highest consumer tier. Here is what happened today, with timestamps so the events speak for themselves.

At 12:54 I sent Anthropic support a detailed email about sustained service quality issues with Claude Code. The email was around 1,500 words and included two evidence attachments totalling 18 pages: a technical incident report on a regression in the latest Claude Code VS Code extension (versions 2.1.141 and 2.1.142), and a candid self-report Claude itself wrote acknowledging a separate session that destroyed a documentation migration on my project. The email cited six prior support tickets, three of which Anthropic's AI agent ('Fin') had explicitly escalated to a human with the message 'we'll email you as soon as an agent has responded.' No human ever replied to any of those three. My email explicitly asked, after six AI-only conversations and three undelivered human-escalation promises, for a response from a human.

At 12:57, three minutes later, I received a reply declining the refund. It is not physically possible for a human to have read either attachment, the email body, the six prior conversation transcripts, or the GitHub issue cluster cited as evidence in three minutes. The reply is signed at the bottom 'This answer was composed by Fin AI Agent, Anthropic's AI Agent.' The Fin signature confirms what the timing already proved.

I replied within the hour, calmly, addressing each point Fin made and giving Anthropic until end of business the following day to put a human on the case before I initiated a chargeback and a complaint to the Irish consumer protection authority. Their response, again within minutes and again from Fin, was to close the ticket entirely: 'This conversation has been closed and is no longer monitored. If you need further assistance, please submit a new request.'

So to be clear about what a premium subscription with Anthropic buys at the moment:

- A technical regression openly documented across 30+ separate user reports on Anthropic's own public GitHub repository, present and unresolved for at least 2.5 weeks. Other users describe it as 'breaks about 50% of my chats or more' and 'I'm not receiving most of my messages.' Anthropic's automated triage bot has been closing these reports as duplicates of an unrelated bug it claims is 'fixed' in the current version. It isn't.

- Documented support failure across multiple contacts. Fin AI handles every interaction. When Fin says a human will follow up, no human does. When the customer escalates with detailed evidence, Fin replies in three minutes and closes the ticket.

- A cancellation flow that doesn't actually let you cancel. My billing settings show only a 'downgrade to Pro' option — there is no cancel button. Under EU consumer law, cancellation must be at least as accessible as subscription. Right now it isn't.

- Extra-usage charges burning through the monthly limit on retries and re-prompts caused by the broken service. I've spent roughly 60% more than the headline subscription rate this cycle and the bulk of it has gone to redoing work the service failed to deliver the first time.

The product itself, when it works, is genuinely the best AI assistant on the market. I have defended it in conversations with other developers. But the gap between the quality of the model and the quality of the support is now wide enough that I cannot in good conscience recommend it to anyone. The 1.4/5 Trustpilot rating across hundreds of reviews exists because this is the standard experience, not the exception.

If you are considering subscribing to Claude Pro or Pro Max, please read the existing reviews carefully before committing your card details. The product is excellent. The support is non-existent. The cancellation path may not work when you need it. And if you have a problem with billing, technical service, or anything else, you will be talking to Fin AI, and Fin AI will not help you.

Paid $20 for a gift subscription, recipient never received it, support bot closes tickets without resolving. No refund issued. Terrible customer experience

OMG... STICK WITH THE FREE SERVICE ITS 10X BETTER THAN THE PAID....
I WAS HITING LIMITS WITH THE FREE SERVICE, SO I PAID THE UPGRADE AND NOW I CANT USE IT AT ALL, KEEPS TELLING ME MY LIMITS ARE UP, PAID THE $21 AND HAVEN'T BEEN ABLE TO USE IT SINCE...
WHAT A LOAD OF CRAP...
GUESS ITS BACK TO DEEPSEEK... (Yes I know its Chinese, but at least it works, you cannot say that about, Claude, ChatGPT, Perplexity, Gemini and OMG GROK

I healed from C-PTSD using Claude. Obviously, I had to do the work, while it provided mirroring. But it was damn f. good at it.

Then i built a framework for truth - found out everything that's wrong wrong with christianity, but right with God.

Again opus 4.7 offered second to none assistance. Incredible to work with.