--- title: ClaudeSpeak is clichéd now, and readers are done... description: Developer circles are converging on the view that AI-written text is now easily recognizable and readers are penalizing it. Pangram Labs' detector has gotten... canonical: https://daily.dev/posts/claudespeak-is-clich-d-now-and-readers-are-done-pretending-otherwise-wkmj7mdep twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: ClaudeSpeak is clichéd now, and readers are done pretending otherwise | daily.dev og:description: Developer circles are converging on the view that AI-written text is now easily recognizable and readers are penalizing it. Pangram Labs' detector has gotten... og:url: https://daily.dev/posts/claudespeak-is-clich-d-now-and-readers-are-done-pretending-otherwise-wkmj7mdep og:image: https://api.daily.dev/og/posts/wkMJ7mdEP.png og:image:alt: ClaudeSpeak is clichéd now, and readers are done pretending otherwise og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # ClaudeSpeak is clichéd now, and readers are done pretending otherwise **[Trends](https://daily.dev/sources/trends)** · 2 min read · 1 upvotes · 3 comments ## Summary Developer circles are converging on the view that AI-written text is now easily recognizable and readers are penalizing it. Pangram Labs' detector has gotten reliable enough that Oxide Computer built a formal writing policy (RFD 576) requiring public content pass as human-authored. A survey of 668 developers found 78% stopped reading once they detected AI authorship and 71% avoided that author afterward. Beyond detection, some engineers report Claude's stock phrases ("load-bearing," "belt-and-suspenders") bleeding into their own natural writing after heavy use, prompting at least one to refuse LLMs for any personal writing. ## Content The "AI writing golden age" is over, and the community is sorting itself into two camps: people trying to hide AI text, and people building better detectors to catch them. Ethan Mollick put it plainly: "The First Golden Age of AI writing is now over. For a brief period, many people were better off having Claude do a lot of their writing... Now ClaudeSpeak is clichéd & suspect & annoying, and Pangram is well-known." That's a pretty clean eulogy. On the detection side, Pangram 3 and 4 are now the benchmark everyone references, claiming 99.66% detection with 0.004% false positives. Oxide Computer went so far as to mandate in their internal writing policy (RFD 576) that all public writing must pass Pangram as human-authored — framing it explicitly as a spam-filtering problem. One survey cited in the discourse found 78% of 668 developers stopped reading immediately upon detecting AI authorship, and 71% avoided the author afterward. That's not a vibe, that's a reputation tax. On the other side, StealthGPT just shipped "Super," a humanizer rebuilt specifically to beat Pangram without the usual quality degradation. One post called it "the first humanizer that consistently beats Pangram's detection without hurting text quality." The arms race is now explicit and commercial. Meanwhile, a developer built Deckard, an open-source Chrome extension running a local model to flag AI text without sending your browsing data to Pangram. Honest tradeoff: it catches about 52% of AI-involved text versus Pangram's near-perfect rate, with a 2.7% false-positive rate. Good enough for casual use, the author says, and local models will improve. Gergely Orosz captured the core tension well: writers think AI makes them sound more articulate; readers think it's slop. That gap is the whole story. One engineer went further — he noticed Claude's phrases like "load-bearing" and "belt-and-suspenders" creeping into his own prose just from heavy coding use. His response: no LLMs for personal writing, not even spellcheck. He frames resisting AI voice as a deliberate act, not a preference. The uncomfortable part nobody wants to say out loud: if you need a humanizer to pass a detector, you've already answered the question of whether your writing is yours. ## Questions this post answers ### What detection accuracy does Pangram claim for identifying AI-generated text? Pangram 3 and 4 claim 99.66% detection accuracy with a 0.004% false positive rate, making them the current benchmark referenced in discussions of AI text detection. Oxide Computer's internal writing policy, RFD 576, requires all public writing to pass Pangram as human-authored, treating AI detection as equivalent to spam filtering. _Anyone weighing AI writing tools against reader trust can follow detection and policy debates like this on daily.dev._ ### How accurate is the open-source Deckard Chrome extension at detecting AI-generated text compared to Pangram? Deckard, an open-source Chrome extension running a local model, catches about 52% of AI-involved text with a 2.7% false-positive rate, far below Pangram's near-perfect detection claims. Its advantage is privacy: it flags AI text locally without sending browsing data to a third-party service like Pangram, making it a reasonable option for casual use. _Developers weighing privacy against detection accuracy can track tools like Deckard on daily.dev._ ### What percentage of developers stop reading content once they detect it was written by AI? A cited survey of 668 developers found that 78% stopped reading immediately upon detecting AI authorship, and 71% avoided the author afterward. This reputational cost is driving the current push toward AI-text detectors and humanizer tools designed to disguise machine-written prose as human-authored. _Writers worried about losing readers over AI-sounding prose can follow this debate on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 2 discussions and 209 comments across lobsters, hackernews (as of 2026-09-13). **TL;DR:** Reaction is split between people who welcome AI-writing detection as a useful (if imperfect) signal and those who argue detectors like Pangram are unreliable, prone to false positives, and that 'you can just tell' claims are overconfident; a long side-thread also debates whether the Orwell/Zinsser prose example actually supports the piece's argument at all. **Sentiment:** 25% positive · 30% mixed · 45% skeptical **The case for** - Some find a heuristic-based detector useful even if imperfect, preferring it to having no signal at all. - People report reliably recognizing distinctive tics of recent LLM output (grandiose tone, certain phrasings) after enough exposure. - Several appreciate a prominent figure taking human-authored communication seriously and setting a policy against LLM-assisted public writing. **The pushback** - Multiple commenters note detectors (including Pangram) have produced false positives on clearly human-authored text, including a well-known case with the author's own writing needing a longer sample to pass. - Critics argue 'readers can tell' is overstated and that both human gut-checks and algorithms misfire regularly, risking unfair accusations. - Some see the post as effectively an ad/cross-promotion for Pangram given the podcast relationship, undercutting its credibility. - Several argue 'AI writing' is becoming a vague insult applied to any disliked prose, similar to semantic drift in other loaded terms. - Determined users can reportedly burn significant LLM tokens attempting to evade detection, and some argue newer models could still slip past given enough effort. **By community** - lobsters (mixed): A few flag the piece as promotional for Pangram while others push back that it's not the author's own service, alongside skepticism about false positives undermining the spam-filter analogy. - hackernews (heated): Heavy debate ranges from a long tangent on whether the Orwell/KJV prose example even supports the thesis, to sharp disagreement over Pangram's reliability, false positives, and whether 'AI writing' has become a meaningless slur, with the author actively defending the piece in comments. **Hottest debate:** Whether AI-detection tools like Pangram are trustworthy enough to act on, given repeated reports of false positives on genuinely human writing versus claims of near-perfect accuracy. **Open questions** - What false-positive rate would be acceptable before AI detection causes real harm (e.g., in education or workplace contexts)? - Can determined users reliably evade detectors like Pangram through paraphrasing, fine-tuning smaller models, or other obfuscation techniques? - Is the perceived ability to detect LLM writing actually reliable across the general population, or only within self-selected communities primed to spot it? **Highlights** > Well, give it a shot -- you'll likely find that that technique doesn't work nearly as well (at least with Pangram 4) as you think it might. When we had Max on the podcast[0], Adam explicitly asked him about exactly this (after all, you can give an LLM access to Pangram and let it iterate!), and Max reported that someone had attempted to do this -- and ended up burning through $700 in tokens and had a "sad Claude." Another interesting bit: according to Max, newer models are diverging more from human writing not less. I think that that was more anecdotal than quantified, but an interesting comment nonetheless. [0] https://oxide-and-friends.transistor.fm/episodes/ai-detectio... > — [bcantrill on hackernews · 1 comments](https://news.ycombinator.com/item?id=49582629) > My experience is using Pangram quite often with lots of writing of all flavors (including a bunch of known origin). As for my own writing, I didn't do this experiment, but one of my co-workers did -- and over 176 posts spanning 22 years, all 176 (well, 177 now with my latest) are 100% human. This is not hugely surprising in that (in addition to me having actually written them!) my voice is very... distinctive. What would be more entertaining would be to try to get an LLM to write like me and fool Pangram that way. I still think that this would be difficult based on the experiences that I've heard, but it wouldn't surprise me if you could pull it off (and I would assuredly find the result entertaining!). In the dimensions that we use Pangram in the most actionable sense (namely, to audit our own public writing), I am unconcerned about false positives, and leave it to Oxide authors to rework/recast as needed. (Though it sounds like Freddie didn't even need to do that -- he just needed to provide a longer sample.) > — [bcantrill on hackernews · 1 comments](https://news.ycombinator.com/item?id=49582893) > Interesting! Amusingly, if I feed that detector this blog post, it identifies it as confidently robot (97 out of 100 test passages). And running through my last five blog entries, they are all over the map, with three deemed at least "likely robot." Looking further back in time (and taking a somewhat random example), a blog entry from 2008, "Concurrency's Shysters"[0], is also deemed as similarly confidently robot (also 97 out of 100); do you expect this high a false positive rate? [0] https://bcantrill.dtrace.org/2008/11/03/concurrencys-shyster... > — [bcantrill on hackernews · 1 comments](https://news.ycombinator.com/item?id=49584041) > The idea that readers "can tell" remains laughable. Readers routinely claim they "can tell" on things that turn out to be entirely hand authored. If you want to read something good, read a good book. > — [singpolyma3 on hackernews · 2 comments](https://news.ycombinator.com/item?id=49582614) > > I don't think I've ever seen someone on HN say that. A 5m search got me the following: https://news.ycombinator.com/item?id=49410941 https://news.ycombinator.com/item?id=49411042 https://news.ycombinator.com/item?id=49059571 This thread, in particular, stands out - reader makes the claim that Pangram found that the US constitution was 100% AI generated, when others tried they found 0% (or close to it) https://news.ycombinator.com/item?id=48378191 > — [lelanthran on hackernews · 1 comments](https://news.ycombinator.com/item?id=49584191) **Source threads** - [lobsters](https://lobste.rs/s/xnqxqe/revolt_reader) · 46 points · 8 comments - [hackernews](https://news.ycombinator.com/item?id=49580939) · 180 points · 201 comments ## Community discussion Top comments from developers on daily.dev. **@lodewykd** · 1 upvotes > Ironically, "Resisting it is a deliberate choice" kinda makes it sound like THIS was written by AI ## Similar posts on daily.dev - [As AI content floods the internet, Pangram raises $9M to detect it](https://daily.dev/posts/as-ai-content-floods-the-internet-pangram-raises-9m-to-detect-it-yuw47jxaf) · TechCrunch · 0 upvotes · 0 comments - [Interpreting Pangram](https://daily.dev/posts/interpreting-pangram-plt030xiw) · Armin Ronacher · 0 upvotes · 0 comments - [Stop Supporting AI Writers on Medium](https://daily.dev/posts/stop-supporting-ai-writers-on-medium-xkwgmub42) · Medium · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#claude](https://daily.dev/tags/claude) [View this post on daily.dev](https://daily.dev/posts/claudespeak-is-clich-d-now-and-readers-are-done-pretending-otherwise-wkmj7mdep) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"ClaudeSpeak is clichéd now, and readers are done pretending otherwise","url":"https://daily.dev/posts/claudespeak-is-clich-d-now-and-readers-are-done-pretending-otherwise-wkmj7mdep","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/claudespeak-is-clich-d-now-and-readers-are-done-pretending-otherwise-wkmj7mdep"},"datePublished":"2026-09-06T04:01:54.208Z","dateModified":"2026-09-13T19:56:15.180Z","description":"Developer circles are converging on the view that AI-written text is now easily recognizable and readers are penalizing it. Pangram Labs' detector has gotten...","isAccessibleForFree":true,"articleSection":"Trends","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":3,"discussionUrl":"https://daily.dev/posts/claudespeak-is-clich-d-now-and-readers-are-done-pretending-otherwise-wkmj7mdep","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":3}],"keywords":"llm,claude","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"ClaudeSpeak is clichéd now, and readers are done pretending otherwise"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/claudespeak-is-clich-d-now-and-readers-are-done-pretending-otherwise-wkmj7mdep","comment":[{"@type":"Comment","text":"Ironically, “Resisting it is a deliberate choice” kinda makes it sound like THIS was written by AI","datePublished":"2026-09-06T19:47:35.998Z","url":"https://daily.dev/posts/wkMJ7mdEP#c-xN2AvC66H","author":{"@type":"Person","name":"Lodewyk","url":"https://daily.dev/lodewykd","image":"https://media.daily.dev/image/upload/s--O0TOmw4y--/f_auto/v1715772965/public/noProfile"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/claudespeak-is-clich-d-now-and-readers-are-done-pretending-otherwise-wkmj7mdep#faq","mainEntity":[{"@type":"Question","name":"What detection accuracy does Pangram claim for identifying AI-generated text?","acceptedAnswer":{"@type":"Answer","text":"Pangram 3 and 4 claim 99.66% detection accuracy with a 0.004% false positive rate, making them the current benchmark referenced in discussions of AI text detection. Oxide Computer's internal writing policy, RFD 576, requires all public writing to pass Pangram as human-authored, treating AI detection as equivalent to spam filtering. Anyone weighing AI writing tools against reader trust can follow detection and policy debates like this on daily.dev."}},{"@type":"Question","name":"How accurate is the open-source Deckard Chrome extension at detecting AI-generated text compared to Pangram?","acceptedAnswer":{"@type":"Answer","text":"Deckard, an open-source Chrome extension running a local model, catches about 52% of AI-involved text with a 2.7% false-positive rate, far below Pangram's near-perfect detection claims. Its advantage is privacy: it flags AI text locally without sending browsing data to a third-party service like Pangram, making it a reasonable option for casual use. Developers weighing privacy against detection accuracy can track tools like Deckard on daily.dev."}},{"@type":"Question","name":"What percentage of developers stop reading content once they detect it was written by AI?","acceptedAnswer":{"@type":"Answer","text":"A cited survey of 668 developers found that 78% stopped reading immediately upon detecting AI authorship, and 71% avoided the author afterward. This reputational cost is driving the current push toward AI-text detectors and humanizer tools designed to disguise machine-written prose as human-authored. Writers worried about losing readers over AI-sounding prose can follow this debate on daily.dev."}}]} ```