Stock photography, transcription, copywriting: commoditization in numbers
What a July 2026 assessment of 10,525 companies shows about AI pressure on image, transcription, and writing tools, with concrete examples of what has already happened to named companies.
Three kinds of work used to sell by the unit: a licensed image, a transcribed minute of audio, a paragraph of marketing copy. General-purpose AI models now produce passable versions of all three at near-zero marginal cost. That does not tell you which specific companies are under pressure and how much. Our assessment does. WillItSurvive assessed 10,525 companies in July 2026, and within that assessment the AI-companies vertical tracks 263 firms, including the image generators, transcription engines, and writing tools that sit closest to these commoditized outputs. This article walks through what the numbers say, company by company, without declaring any outcome. Our policy is simple: we publish standings and dated watch-items, and we never call a result in advance.
The category baseline
Of the 263 AI companies assessed in July 2026, 170 were rated losing its protection and only 2 were rated hard to displace. Revenue pressure was elevated across the group: 88 of the 263 companies carried high pressure at the time of assessment. The pattern the vertical shows repeatedly: value that lives above the model layer, as a thin product wrapper, gets absorbed when the model labs ship the same feature natively. Image, transcription, and copywriting tools are the clearest examples, because their entire output is exactly what a general model produces.
Images: the generators are competing with each other and with free
Stock photography's economics changed when image generation became a commodity feature. The assessment does not track legacy stock agencies in this vertical, but it does track the generators reshaping the market, and none of them cleared a standing better than protection under test. Midjourney was rated losing its protection in the July 2026 assessment, Stability AI was rated protection under test, and Ideogram was rated little protection left. Leonardo AI (Protection under test) and Krea AI (Losing its protection) sit in the same band. The relative bright spot is Photoroom, rated protection under test in the July 2026 assessment, a tool aimed at a specific commerce workflow rather than general image generation. Black Forest Labs, a model builder rather than a wrapper, was rated losing its protection. The takeaway: even the companies doing the disrupting face heavy pressure from the next model release, because the output itself is interchangeable.
Transcription: high pressure on the consumer end, moderate on infrastructure
Transcription splits into consumer-facing note-takers and developer-facing speech APIs, and the July 2026 standings split the same way. Otter.ai, the best-known consumer transcription brand, was rated losing its protection in the July 2026 assessment. Meeting transcription is now a native feature of Zoom, Teams, and Google Meet, which is exactly the bundling dynamic the vertical's pressure map documents across categories. The infrastructure side fares somewhat better on paper: Deepgram and AssemblyAI were also rated losing its protection, but they sell to developers building products rather than competing head-on with a bundled meeting feature. Adjacent audio tools cluster in the same band: Descript, Speechify, and ElevenLabs were all rated losing its protection as of July 2026. Nobody in the speech-and-audio cluster reached a standing above losing its protection.
Copywriting: the most exposed cluster in the vertical
Copywriting tools were the first visible AI-wrapper business, and they now sit among the most exposed companies in the assessment. Copy.ai, Jasper AI, and Grammarly were all rated losing its protection in the July 2026 assessment. Writer, which targets enterprise workflows rather than one-off copy, carried the same standing. DeepL, whose translation product overlaps heavily with what general models now do free, was rated losing its protection as well. The mechanism is documented in the vertical's own sector analysis: ChatGPT, Claude, and Gemini produce marketing copy, grammar correction, and translation as default behavior, so a standalone subscription has to justify itself on workflow, brand controls, or integration rather than on the text itself.
What already happened to thin layers
The vertical's records include companies that were absorbed rather than outcompeted. Phind, whose developer search was subsumed by native search in ChatGPT, Claude, and Gemini, shut down in January 2026 and was rated little protection left in the July 2026 assessment. Humanloop's team was absorbed by Anthropic. Rewind AI was acquired and shut down by Meta. Covariant became a shell after Amazon hired its founders in 2024. All three carried a little protection left standing at the time of assessment. These are the closed cases the open cases should be read against.
Questions this data answers
Is copywriting software the most AI-pressured category? It's among the most pressured groups in this assessment. Copy.ai, Jasper AI, and Grammarly were all rated losing its protection in the July 2026 assessment, and each competes directly with ChatGPT, Claude, and Gemini generating the same copy, grammar correction, and translation output for free.
Are transcription companies in the same position? The consumer end is: Otter.ai was rated losing its protection in July 2026, with meeting platforms like Zoom, Teams, and Google Meet now bundling transcription natively. The API providers Deepgram and AssemblyAI carry the same standing but sell to developers rather than compete head-on with a bundled meeting feature.
Do the image-generation companies themselves face commoditization? Yes. Midjourney, Stability AI, and Ideogram were all rated losing its protection or worse in the July 2026 assessment, and none held a standing better than protection under test.
Which AI companies avoided this squeeze? Of the 263 assessed in July 2026, only Applied Intuition and Dataiku were rated hard to displace. The vertical's broader pattern favors chips, physical infrastructure, and exclusive data over products layered on someone else's model.
Watch-items, dated
By mid-2027, check whether more wrapper startups exit via acquihire rather than new funding rounds, and whether Otter.ai, Jasper, or Grammarly announce repositioning toward enterprise workflow rather than raw output. Both are checkable signals, not conclusions.
Assessments reflect evidence available July 2026. Companies change; treat watch-items as things to verify, not conclusions.
Figures are as of each company's assessment date unless dated otherwise.