{"slug": "openai-has-commanding-lead-geminis-usage-has-fallen-shows-data-from-ai-text", "title": "OpenAI Has Commanding Lead, Gemini’s Usage has Fallen, Shows Data From AI Text Detector Pangram", "summary": "OpenAI's models have held at least 50% of AI-generated text market share every month since mid-2024, reaching 74.7% by July 2026, according to data from AI text detector Pangram. Anthropic's share climbed steadily from 4.3% to 14.9% over the same period, while Google's Gemini fell to 1.9%, contrasting with traffic-based trackers like SimilarWeb that show Gemini gaining ground.", "body_md": "Different data sources are giving different — and interesting — insights about the AI race.\n\nThe latest comes from Pangram, an AI text detection company that says it can identify which model family generated a given piece of writing. Pangram built its business around catching AI-generated text for schools and publishers, but the same underlying classifier can be pointed at a different question: if you sample enough text from across the internet every month, what does the mix of models producing that text say about who’s actually winning the AI race.\n\nPangram’s charts, drawn from anonymized samples pulled from its dashboard since mid-2024, tell a story that looks slightly different to the one [SimilarWeb’s traffic data](https://officechai.com/ai/chatgpt-just-about-clings-on-to-50-market-share-in-may-data-gemini-claude-gain/) has been telling for the past year.\n\n## OpenAI’s Dominance, By Text Volume\n\nAccording to Pangram, OpenAI’s models have not dropped below 50% market share in a single month since the company started tracking this data. As of July 2026, OpenAI sits at 74.7% of all AI-generated text sampled, and the chart shows that figure holding remarkably steady through most of 2025 and 2026, rarely dipping below 70% even during the months when Gemini and Claude were making their most aggressive plays.\n\nThat number sits in sharp contrast to what the web traffic trackers have been reporting. SimilarWeb’s most recent numbers put ChatGPT’s share of generative AI web traffic at around 52%, a figure that has been sliding for over a year now, down from more than 76% just twelve months prior. Pangram’s method isn’t counting visits or sessions, though — it’s sampling actual written output and guessing which model produced it, which naturally weights the numbers toward whichever model is generating the most text in the wild, whether that’s through chatbot conversations, API calls embedded in other products, or content that eventually ends up published somewhere.\n\nThat distinction matters a lot here. OpenAI’s models power an enormous amount of downstream content generation through API integrations that never show up in a website traffic count, and Pangram’s methodology would presumably capture a lot of that.\n\n## Anthropic Quietly Building Share\n\nThe second big trend in Pangram’s data is Anthropic’s steady climb, from 4.3% when the company started tracking in mid-2024 to 14.9% as of July 2026. It’s a gradual, near-uninterrupted rise across two years of monthly data, without the kind of sharp single-month jump you’d expect from a viral product launch.\n\nThis tracks with a broader pattern showing up across several other data sources this year. Ramp’s payments data found Anthropic’s [share of combined OpenAI-Anthropic business spend](https://officechai.com/ai/anthropic-has-passed-openai-in-us-business-adoption-for-the-first-time-says-ramp-data/) jumping from around 10% at the start of 2025 to over 65% by this February, and Menlo Ventures has separately reported Anthropic overtaking OpenAI in enterprise LLM API usage. On the consumer side, Claude’s traffic share on SimilarWeb went from under 2% a year ago to over 9% by June 2026, the sharpest sustained climb of any platform tracked over that stretch.\n\nWhat’s notable is that Pangram’s number, sourced from an entirely different methodology, lands Anthropic’s growth in a similar zone — climbing steadily, gaining ground every few months, without ever approaching a majority share.\n\n## The Gemini Numbers\n\nThis is where Pangram’s data diverges most sharply from everything else being reported. The chart shows Google’s models falling from around 12% in the platform’s early days to just 1.9% by July 2026.\n\nThat’s a dramatically different picture from what traffic and usage trackers have shown all year. SimilarWeb had Gemini crossing 25% of generative AI web traffic back in March, and Pew Research’s February survey found that [44% of Americans say they use ChatGPT](https://officechai.com/ai/44-of-americans-say-they-use-chatgpt-compared-to-24-for-gemini-pew-report/) compared to 24% for Gemini — putting Google comfortably in second place, well ahead of Anthropic’s 6% in that same survey. None of the major traffic or usage datasets from this year put Gemini anywhere close to a 1.9% share of anything.\n\nThe gap likely comes down to what Pangram is actually measuring. A text classifier trained to spot the stylistic fingerprints of AI writing is going to be most reliable at flagging text that gets published somewhere — essays, articles, documents, the kind of content that ends up in the training and sampling pipelines these detectors draw from. Gemini had first gone viral for its image generation capabilities through Nano Banana, and could be being used more for personal queries than for professionally-created text. As such, these numbers could be lower for Gemini than its web traffic would suggest.\n\nAlso, Pangram itself flags this limitation upfront, noting its classifier is 91% accurate at the individual document level, which isn’t reliable for predicting any single piece of text but becomes more useful when looking at trends across large volumes of samples over time.\n\n## Reading The Numbers Together\n\nNone of this means one dataset is right and the others are wrong. Traffic trackers measure where people go, survey data measures what people say they use, payments data measures what businesses actually pay for, and Pangram is measuring something else entirely: the composition of AI-generated text circulating in whatever corpus it samples from. Each lens picks up a different slice of a market that has gotten considerably more fragmented and harder to summarize in a single number than it was even a year ago.\n\nWhat is consistent across nearly every dataset, Pangram’s included, is that ChatGPT is the current market leader, and Anthropic is gaining ground. And with Gemini’s traffic share [having fallen](https://officechai.com/ai/geminis-traffic-share-falls-for-first-time-in-over-a-year-as-its-growth-over-chatgpt-plateaus/) for the first time in over a year, and with Pangram also saying that Gemni is losing ground, Google would need to lift its game to continue to be competitive in the AI space.", "url": "https://wpnews.pro/news/openai-has-commanding-lead-geminis-usage-has-fallen-shows-data-from-ai-text", "canonical_source": "https://officechai.com/ai/openai-has-commanding-lead-geminis-usage-has-fallen-shows-data-from-ai-text-detector-pangram/", "published_at": "2026-08-12 08:33:42+00:00", "updated_at": "2026-08-12 08:47:44.513408+00:00", "lang": "en", "topics": ["artificial-intelligence", "generative-ai", "ai-tools"], "entities": ["OpenAI", "Pangram", "Anthropic", "Google", "Gemini", "SimilarWeb", "ChatGPT", "Claude"], "alternates": {"html": "https://wpnews.pro/news/openai-has-commanding-lead-geminis-usage-has-fallen-shows-data-from-ai-text", "markdown": "https://wpnews.pro/news/openai-has-commanding-lead-geminis-usage-has-fallen-shows-data-from-ai-text.md", "text": "https://wpnews.pro/news/openai-has-commanding-lead-geminis-usage-has-fallen-shows-data-from-ai-text.txt", "jsonld": "https://wpnews.pro/news/openai-has-commanding-lead-geminis-usage-has-fallen-shows-data-from-ai-text.jsonld"}}