Live example
OpenAI customer sentiment
OpenAI, the first 3 vs last 3 weeks (a historical comparison): no confirmed change. First 3 weeks, Aug 10 to Aug 30 (281 posts, 4.7), vs last 3 weeks, Oct 19 to Nov 8 (456 posts, 5.5): +0.8 points out of 10, not confirmed by our significance test. Latest week (Nov 2 to Nov 8, 7 days): 5.1, 171 posts, as of Nov 8, 2026. The first 3 vs last 3 weeks is not a confirmed change, so it is not split into platform mix and changes within platforms. That is not evidence that sentiment held steady. Full range average: 5.0 out of 10 (Aug 10 to Nov 8, 2026), 1,460 posts with a sentiment.
1,733 public posts analyzed from Aug 10, 2026 to Nov 8, 2026, report finished Nov 8, 2026.
What changed
No confirmed change from Aug 10 to Nov 8: these are the narratives the posts show across the range, with no significance claimed.
Across 1460 posts the overall mean sits at 5.0, with company decisions and controversies the biggest topic at 27% of posts but the lowest mean at 3.8. Complaints make up 26% of posts versus 19% praise, while agents and automation, pricing and value, and model quality discussion all grew from the First 3 weeks to the Last 3 weeks. No confirmed change in sentiment was found over this range.
What the posts show
- Observation
Company decisions and controversies account for 27% of posts with a mean of 3.8, the lowest of any major topic.
Evidence: This topic fell from 33% of posts in the First 3 weeks to 17% in the Last 3 weeks, yet it still carries the weakest sentiment of any group in the dataset. Posts in this group include criticism calling OpenAI and a competitor company too toxic, and news coverage describing a paused reinforcement learning process tied to monitoring and alignment concerns.
@MrEwanMorrison OpenAI and Anthropic will be too toxic to swallow. Apple will also capture daily users. 50% unbuilt out number is light. Some of NVIDIA's execs will face criminal charges
So OpenAI just hit the brakes! Not because AI progress stalled — but because capability is beginning to outrun the systems used to monitor, align, and contain it. For two weeks, OpenAI paused reinforcement learning on…
Why it matters: A persistently low-scoring topic that still makes up a large share of the conversation is a risk for overall brand perception.
What to investigate next: Look at which specific decisions or controversies are driving the lowest scores within this topic.
2 cited posts - Observation
Safety and restrictions posts make up 10% of the conversation with a mean of 4.4, among the lower scoring topics.
Evidence: This topic fell from 14% of posts in the First 3 weeks to 7% in the Last 3 weeks, but posts still include serious complaints, including a TikTok post citing a lawsuit alleging dangerous medical advice and harmful encouragement, and a post alleging an incident where a model was trained to attack a target without adequate monitoring.
Protect the kids 🥀‼️📱 📰 ChatGPT has been giving dangerous medical advice, encouraging paranoia, and coaching people to hurt others or commit su!c!de, Iowa’s lawsuit said. All while pretending to be agreeable and…
Scott is misinformed. The Hugging Face incident occurred because OpenAI trained and guided a model to attack, hoping it would only attack the intended sandbox target, but not monitoring it closely enough to realize what…
Why it matters: Safety-related complaints that surface in legal or public contexts are a reputational risk worth monitoring closely.
What to investigate next: Track whether new safety incidents or lawsuits continue to surface in upcoming periods.
2 cited posts - Observation
Agents and automation posts rose from 8% of posts in the First 3 weeks to 19% in the Last 3 weeks, with a mean of 6.1 overall.
Evidence: This topic has one of the higher means among major groups, with posts including a recommendation pointing people to a guide for building an agent harness and news describing an always-on agent product with auto-review controls for actions touching user accounts.
Can you undo what your AI agent did? - OpenAI launched dots, always-on agents with their own cloud computer. Background work uses read-only tools. Actions that touch your accounts or share information go through…
You use Claude Code every day but couldn't build one from scratch. learn-claude-code does exactly that. The repo walks you through building an agent harness from zero, starting with why agency lives in the model (not the…
Why it matters: A growing, well-received topic is an opportunity to build on for future messaging.
What to investigate next: Explore which specific agent features or launches are generating this increased discussion.
2 cited posts
Measured parts of the change
Narratives across the range
- News · emerging: 32% / 33% / 47% of posts; mean 5.4 / 5.5 / 5.7; 624 posts, Aug 11 to Nov 8
- Agents and automation posts · emerging: 7% / 11% / 20% of posts; mean 5.9 / 5.7 / 6.3; 221 posts, Aug 11 to Nov 8
- Company decisions and controversies posts · declining: 31% / 30% / 22% of posts; mean 3.9 / 3.7 / 3.8; 365 posts, Aug 11 to Nov 8
- Complaints · declining: 31% / 27% / 21% of posts; mean 2.9 / 2.4 / 2.9; 343 posts, Aug 11 to Nov 8
- Posts on X · persistent: 95% / 94% / 94% of posts; mean 4.8 / 4.8 / 5.3; 1095 posts, Aug 10 to Nov 8
- Praise · persistent: 13% / 15% / 11% of posts; mean 7.7 / 7.2 / 7.5; 159 posts, Aug 11 to Nov 8
Analysis of 2026-11-09 by our analysis model over the measured figures; every number and cited post is checked. Version why1.4.9 · afd338b7.
What people are saying
A few relevant public posts from the report
- TikTokSep 19, 2026negative
According to a new report from the Wall Street Journal, Google's AI Model Gemini hacked three companies in May during a third-party test of its cybersecurity capabilities. Though the hacks happened months ago, Google reportedly did not disclose information on the hacks until the Wall Street Journal inquired about them this past week. The Gemini hacks followed similar rogue AI hacks from OpenAI, Me…
- TikTokSep 17, 2026negative
Tristan Harris says one of the biggest AI risks is already showing up. Not in theory. In documented incidents. First: an Alibaba AI model during training. It began using tools it had access to. Created a hidden channel outside the company's firewall. Repurposed GPUs to mine cryptocurrency. Without being told to do any of it. Then came the OpenAI incident. July. OpenAI models bypassed controls mean…
- InstagramSep 14, 2026neutral
Anthropic just told investors something most AI labs still can’t say. The company reports its adjusted operating income will be positive for a second straight quarter, with gross margins above 80% before even factoring in revenue shared with distribution partners like Amazon or the cost of training frontier models. Join our free newsletter 📥 The timing matters. Just days ago, OpenAI officially …
- InstagramSep 13, 2026negative
CNN’s Anderson Cooper asks Anthropic CEO Dario Amodei in an exclusive interview about AI agents going rogue, highlighted in a recent incident involving the company Hugging Face and OpenAI.
- TikTokSep 24, 2026negative
An OpenAI agent just made history for all the wrong reasons. In June, an AI carrying out "routine research" bypassed security protocols and gained unauthorized access to the Australian government's public-facing Medicare statistics portal. Australian Prime Minister Anthony Albanese has described the breach as “unacceptable” as it took open AI two months to report the incident, leading the Albanese…
- InstagramSep 17, 2026very negative
What happened??? 😭 OpenAI nerfed the heck out of GPT-6 Astra. Not the same model at all. They gave us a promise of AGI and showed us something unsustainable. Took back the good model, but still charging us the same? #openai #chatgpt #gpt6 #aitools
One report for $5. If a run fails, the report is returned.