Which AI tools are people canceling now?

Last updated: 17 September 2026

SUMMARY

ChatGPT Plus probably sees the largest number of consumer AI cancellations today, while Midjourney, generic AI writing tools, basic research products and simple meeting notetakers face the strongest structural pressure.

AI subscriptions have a real retention problem. RevenueCat finds that AI apps monetize payers unusually well at first, but churn about 30% faster than non-AI subscription apps and retain fewer monthly and annual subscribers after a year.

That does not mean people are abandoning AI. Consumer AI spending has risen sharply at the same time, which points to a market where users are adding, testing, canceling and replacing subscriptions much more freely than they do with traditional software.

ChatGPT is the clearest example of why cancellation volume and product weakness are not the same thing. It tops the strongest direct cancellation-intent survey we found, yet its enormous paid base, usage and revenue mean even heavy churn can coexist with a very strong underlying business.

Midjourney looks more exposed. Its high unused-subscription rate coincides with a long-term fall in its relative web ranking, while image generation has become a standard feature inside ChatGPT, Gemini, Canva and other broader products.

Overlap is becoming a bigger problem than price by itself. Paying $20 or $30 for one useful AI tool is manageable; paying separately for an assistant, image generator, research engine and meeting summarizer becomes harder to justify when the products increasingly perform the same jobs.

The most vulnerable standalone tools are those built around a generic capability rather than a workflow. Basic AI writing, rewriting, search, image generation and meeting summaries are all being absorbed into larger platforms that customers already use.

Claude and Gemini show why cancellation data needs context. Both can be cut from a personal subscription stack, yet current paid-user, usage and business-adoption data remain strong, so much of the churn appears to be switching and consolidation rather than abandonment.

AI coding products look stickier for now. Developers may rotate between Cursor, Claude Code, Codex, GitHub Copilot and Gemini, but the category has a much clearer daily-use case and a much easier return-on-investment calculation than occasional writing or image subscriptions.

The broader pattern is simple: people are pruning extra AI subscriptions faster than they are reducing AI usage. Products tied to frequent work, proprietary context or a wider software ecosystem are much harder to remove than tools whose main feature can be recreated elsewhere in a few clicks.

Get the biggest database of
profitable internet businesses

We mapped 300+ proven digital businesses so you can skip the blind trial and error. For each one, you get the site, the revenue numbers, the distribution strategy, the repeatable patterns, and ideas to recreate the model in a different niche, channel, or angle.

Get the full database →

Are AI subscriptions actually being canceled more now?

Yes. AI apps currently have a clear retention problem, even though people are spending far more money on AI than a year ago.

RevenueCat’s latest subscription benchmark covers more than 115,000 apps and over $16 billion in revenue. AI apps make more money from each payer than traditional subscription apps, but they lose those payers faster. After 12 months, median retention for monthly AI subscriptions is 6.1%, compared with 9.5% for non-AI apps. Annual AI plans retain 21.1%, versus 30.7% for other apps. RevenueCat summarizes the overall gap more simply: AI apps churn about 30% faster.

The difference is large enough to rule out the idea that we are simply seeing normal subscription churn dressed up as an AI story. AI is unusually good at getting people to pay, then unusually weak at keeping them.

At the same time, demand for AI keeps growing. Menlo Ventures’ latest survey of 5,067 U.S. adults estimates that global consumer AI spending has climbed from roughly $12 billion to $40 billion in one year. Fifty-five percent of AI users now say they pay for at least one AI product, while 46% of existing payers say they spend more than they did a year ago.

People are spending more on AI overall while canceling individual subscriptions much more freely. They try products, keep the ones that become habits and cut the ones that overlap with something else.

Which AI tools do people say they are most likely to cancel now?

ChatGPT currently produces the biggest cancellation signal in the cleanest product-level survey we found, while Claude and Gemini come next.

Self Financial surveyed 1,272 U.S. adults about paid subscriptions and asked paid AI users both what they had stopped using and what they were considering canceling. Among respondents who said they would unsubscribe from an AI service, 46.3% named ChatGPT, 36.6% named Claude and 29.9% named Gemini.

The unused-subscription numbers are even more interesting. Half of surveyed ChatGPT subscribers said they had paid for ChatGPT without using it for at least 30 days. Midjourney came next at 42.6%, followed by Canva AI at 40.2%. Jasper, Grammarly, Notion AI, Writesonic, Perplexity and Gemini all landed in the mid-to-high 30s. Claude stood out at just 27.2%.

We should be careful with these figures. Self does not publish the number of respondents behind every individual product percentage, so the smaller services may have much wider sampling error. Saying that 36.6% of cancellation-minded Claude users chose Claude is very different from proving that 36.6% of all Claude subscribers cancel.

Still, this is the clearest direct evidence we have of where consumers themselves are thinking about cutting AI spending.

AI tool Paid subscription unused for 30+ days Cancellation signal
ChatGPT 50.4% Highest stated cancellation intent
Midjourney 42.6% Very high unused rate
Canva AI 40.2% High unused rate
Grammarly AI 38.7% High unused rate
Jasper AI 37.8% High unused rate
Notion AI 36.8% High unused rate
Writesonic 36.6% High unused rate
Perplexity 35.8% High unused rate
Gemini 35.3% Third-highest stated cancellation intent
Claude 27.2% Second-highest stated cancellation intent, but lowest unused rate

Building a digital business?

We have mapped 300+ proven internet businesses. You'll get the full breakdown: revenue, distribution, why it works and how to replicate.

GET THE FULL DATABASE → $49

Does ChatGPT really have a cancellation problem today?

ChatGPT probably generates more consumer cancellations than any other AI tool today, largely because its paid user base is huge and downgrading is easy.

The Self Financial survey looks rough for ChatGPT: 50.4% of the surveyed paid users had gone at least 30 days without using their subscription, and ChatGPT was the service most frequently named by people considering an AI cancellation.

Yet current usage data gives us little reason to think ChatGPT itself is being abandoned.

Sensor Tower says ChatGPT’s mobile active user base approached one billion earlier this year. Its mobile revenue reached nearly $1.3 billion in a single quarter, more than five times the combined revenue of the other top-ten AI apps in Sensor Tower’s comparison. Andreessen Horowitz also reported, using Yipit data, that ChatGPT still had around eight times as many U.S. paid subscribers as Claude and four times as many as Gemini at the beginning of the year.

ChatGPT also gives users several cheaper ways to stay. Plus still costs $20 a month, while ChatGPT Free remains useful enough for casual users and ChatGPT Go has expanded globally at a lower price point. A user can therefore stop paying $20 without giving up ChatGPT.

Millions of people can subscribe for a demanding project, use the product heavily for a few weeks, then fall back to Free or Go. ChatGPT Plus can be heavily canceled without ChatGPT itself being in trouble.

Are Claude Pro users canceling Claude now?

Claude has visible cancellation intent, but current usage and payment data still point toward growth rather than broad subscriber flight.

The contrast inside the Self Financial survey is striking. Claude ranked second when respondents were asked which AI subscription they would consider canceling, yet Claude had the lowest 30-day unused rate of all ten AI products measured.

Other datasets make the picture clearer. Andreessen Horowitz reported that Claude’s U.S. paid subscriber count was growing more than 200% year over year at the beginning of the year, based on Yipit data. The same analysis described Claude and ChatGPT as having some of the strongest consumer paid-subscriber retention in the category.

Claude is also gaining ground among companies. Ramp’s newest vendor data shows Anthropic being used by 80% of Ramp customers that purchased from a foundational-model provider over the previous 12 months, up sharply year over year. OpenAI stood at 77% on the same measure. Ramp also found that 66% of companies entering the category chose Anthropic, versus 39% for OpenAI.

Those numbers measure businesses rather than Claude Pro consumers, so we should not pretend they are the same thing. They do show that Claude’s wider product ecosystem is currently getting stronger.

Claude users are clearly willing to switch plans and providers, especially because ChatGPT, Gemini and Claude overlap so heavily. The harder claim — that people are broadly giving up on Claude — does not fit the latest evidence.

Stop testing random ideas

Start from proof. 300+ profitable internet businesses, mapped, broken down, and ready to copy, in one searchable database.

STEAL WHAT WORKS → $49

Is Gemini getting canceled because Google gives AI away everywhere?

Gemini subscriptions are exposed to cancellation, while Gemini usage itself is currently growing extremely fast.

This is one of the strangest cases in the market because Google keeps making the standalone subscription less central to the Gemini experience.

Sensor Tower found Gemini gaining sharply against ChatGPT in web usage: among leading AI platforms, Gemini’s share of visits rose to 22% while ChatGPT’s fell from 66% to 50% between mid-2025 and early 2026. Menlo Ventures’ latest consumer study found Gemini used by 58% of surveyed AI users, barely behind ChatGPT at 60%.

Google can reach those users through far more than gemini.google.com. Gemini now sits inside Search, Android, Chrome, Gmail, Docs, Sheets, Meet and other Google products. Someone may therefore stop seeing a separate Gemini payment as essential while using Google AI more often than before.

Paid Gemini is growing in some datasets too. Andreessen Horowitz reported that U.S. Gemini paid subscriptions were up 258% year over year at the start of the year.

Gemini has enough overlap with other assistants to get cut from an individual’s subscription stack, especially by someone already paying for ChatGPT or Claude. Google’s distribution keeps bringing that same person back into Gemini through products they already use.

A canceled Gemini subscription can therefore tell us surprisingly little about whether that user is moving away from Google AI.

Is Midjourney one of the AI tools people are really cutting now?

Yes. Midjourney has one of the strongest structural cancellation cases among major AI products today because general-purpose assistants have absorbed much of casual image generation.

As seen above, 42.6% of paid Midjourney users in the Self Financial survey had left their subscription unused for at least 30 days. That figure becomes much more meaningful once we connect it with the category’s longer trajectory.

Andreessen Horowitz has tracked consumer AI products since 2023. Midjourney ranked inside the top ten web AI products in its first edition. In the latest edition, it ranked 46th.

The whole image-generation category has compressed in the same way. Seven of the nine creative AI products in a16z’s first web ranking were image generators. Only three standalone image generators remained three years later. Video, music and voice products took many of the places image tools lost.

ChatGPT and Gemini explain a lot of that change. Both now generate and edit images inside products people already use for writing, research, search, files and coding. Canva does the same inside a full creative suite.

Midjourney keeps improving — V8.2 became its default model recently, with better image quality, personalization and editing — and serious visual creators still have reasons to pay. Standard currently costs $30 a month, with higher tiers at $60 and $120.

The pressure comes from occasional users. Someone generating a handful of illustrations, marketing visuals or concept images every month can now get acceptable results inside another subscription.

Among big AI brands, Midjourney is one of the clearest cases where cancellation pressure reflects a deeper change in what customers actually need.

Looking for a profitable business idea?

Get our database of 300+ profitable internet businesses, mapped, broken down, and ready to copy.

STEAL WHAT WORKS → $49

Which standalone AI tools are easiest to cancel these days?

Generic writing, image generation and basic research tools are becoming especially easy to cut because their main features increasingly exist inside larger AI products.

Consider what has happened to AI writing. Jasper and Writesonic originally benefited from a clear reason to buy a separate product: they packaged generative writing before general assistants became good enough for everyday marketing copy. Now ChatGPT, Claude and Gemini can write product descriptions, blog drafts, emails, ad copy and social posts in the same subscription people use for many other jobs.

The two companies have reacted in ways that tell us a lot about the market. Jasper has moved deeper into enterprise marketing workflows, brand control and campaign tools. Writesonic increasingly sells SEO, AI-search visibility and marketing automation. Both are trying to move further away from simple text generation.

Search has the same problem. Perplexity still has a strong product for people who research heavily, but cited web answers are no longer unusual. ChatGPT, Claude and Gemini all have web and deep-research capabilities now. A casual user who mainly asks Perplexity a few questions each week has more alternatives than two years ago.

Image generation has moved furthest down this path. Midjourney’s relative web ranking has fallen sharply as image creation became a built-in feature elsewhere.

The weak spot across these categories is pretty obvious: the standalone tool has to keep proving why its specialized workflow deserves another monthly bill.

Type of AI tool Examples Why cancellation gets easier
General AI writing Jasper, Writesonic ChatGPT, Claude and Gemini already handle basic writing well
Standalone image generation Midjourney and smaller image tools Image generation is bundled into general assistants and design suites
AI search and research Perplexity and smaller research tools General assistants increasingly offer search, citations and deep research
Basic AI editing Standalone rewrite/paraphrase products Writing suites and chatbots now include the same functions
Generic chatbot wrappers Smaller assistant apps Little product lock-in and many free substitutes

Are Canva AI, Grammarly AI and Notion AI getting canceled too?

Some users are cutting these AI subscriptions, but Canva, Grammarly and Notion have much stronger protection than standalone one-feature AI apps.

The difference comes from the product around the AI.

A Canva customer may use presentations, templates, video editing, brand kits and collaborative design even during a month when generative AI barely gets touched. Grammarly still has a core writing and editing workflow beyond chatting with a model. Notion embeds AI into a workspace that can already contain a company’s notes, databases, projects, meetings and internal knowledge.

Notion provides the clearest example of how powerful that bundling can become. Andreessen Horowitz reported that Notion’s paid AI attach rate jumped from roughly 20% to more than 50% in one year, with AI-related functionality contributing roughly half of company ARR.

Midjourney has to justify Midjourney every month. Notion AI can become one part of why someone keeps Notion.

The same logic increasingly applies to Canva and Grammarly. AI can disappear into the product rather than remaining a separate thing users consciously decide to renew.

We should therefore expect plenty of AI-feature pruning and plan changes around these products, while the wider customer relationship remains harder to break.

Get the biggest database of
profitable internet businesses

We mapped 300+ proven digital businesses so you can skip the blind trial and error. For each one, you get the site, the revenue numbers, the distribution strategy, the repeatable patterns, and ideas to recreate the model in a different niche, channel, or angle.

Get the full database →

Are people canceling AI coding tools like Cursor and Claude Code?

AI coding tools currently look much stickier than generic writing or image subscriptions because developers can use them for hours every day.

The latest usage data still shows strong demand. Cursor remains inside Andreessen Horowitz’s top 50 consumer AI web products. Claude Code reached a $1 billion annualized revenue run rate within six months, according to Anthropic figures cited by a16z. OpenAI reported roughly two million weekly Codex users earlier this year, with usage still growing quickly at that point.

A coding subscription also has an easier value calculation. If a developer uses Cursor, Claude Code or Codex every working day and saves meaningful time, $20 or $50 a month can feel trivial relative to the value of that time.

Competition is intense, though. Developers can move between Cursor, Claude Code, Codex, GitHub Copilot, Gemini and direct model APIs. They often test several at once. That creates plenty of churn at the vendor level.

The category itself still looks healthy. Andreessen Horowitz says traffic across the leading vibe-coding platforms has continued to grow, even after the initial explosion slowed, while revenue has kept rising for several products.

So when a developer cancels one coding AI today, there is a good chance the money simply moves to another coding AI.

Are AI meeting notetakers becoming easy to cancel?

Yes, basic AI meeting notetakers are getting easier to cut because Google, Microsoft, Zoom and Notion increasingly include the same core functions inside software companies already buy.

Fireflies, Fathom, Otter, TL;DV and Granola built strong businesses by solving an obvious problem: recording meetings, transcribing them, summarizing what happened and pulling out action items.

That gap is closing quickly.

Google has Gemini across Meet and Workspace. Microsoft puts Copilot into Teams and Microsoft 365. Zoom has AI Companion. Notion now offers AI meeting notes inside the same workspace teams use for documents and projects.

Dedicated meeting tools still have meaningful reach. Andreessen Horowitz estimates that Fireflies, Fathom, Otter, TL;DV and Granola collectively generate around 20 million web visits. Specialized products can also offer better cross-platform support, sales integrations, searchable meeting history and conversation intelligence.

The weakest subscription is the one being used only for transcripts and summaries. Once a company realizes that capability is already inside its Microsoft, Google, Zoom or Notion bill, the separate expense becomes very easy to question.

This category should split. Deep meeting workflows can survive as specialized software. Plain transcription and summarization will have a much harder time staying separate.

Building a digital business?

We have mapped 300+ proven internet businesses. You'll get the full breakdown: revenue, distribution, why it works and how to replicate.

GET THE FULL DATABASE → $49

Why do people cancel paid AI tools so quickly?

AI subscriptions often solve an urgent problem immediately, then struggle to become something the customer needs every month.

RevenueCat’s retention curve makes that visible. AI apps convert free trials to paid subscriptions 52% better than non-AI apps at the median, and they generate 41% more realized first-year revenue per payer. Yet their long-term retention is considerably worse.

The first renewal is where much of the damage happens.

That makes sense when we look at how people buy these products. Someone can subscribe to an image generator because they need visuals for one launch. Another person pays for ChatGPT Plus during an intensive research project. A student upgrades around exams. A marketer buys an AI writing product while producing a batch of content. Once the immediate workload disappears, the subscription has to earn its place again.

Free tiers make that decision even easier. ChatGPT, Claude, Gemini and many smaller AI products let users keep using a weaker or more limited version after they stop paying.

Switching costs are also unusually low. A CRM may contain years of customer records, workflows and staff training. Moving from Claude to ChatGPT can take seconds.

Ramp sees the same fluidity among businesses. In its summer spending data, 52% of companies that used either Anthropic or OpenAI were paying both. The category still has plenty of customers testing one provider beside another rather than committing permanently.

AI subscriptions face a brutal monthly question: “Did I use this enough to pay again?” Products without a strong answer lose users fast.

Is price really why people cancel AI tools, or is overlap the bigger problem?

Overlap looks like the deeper problem, with price acting as the trigger that forces users to deal with it.

A single mainstream AI subscription remains relatively affordable. ChatGPT Plus costs $20 a month. Midjourney Standard costs $30. Similar paid plans across assistants, search tools, coding products and creative apps quickly turn that into a much larger bill.

Menlo Ventures found that the typical paying AI user spends $20 to $49 per month. Only 14% of payers spend at least $100 monthly, although that small group generates around 60% of all consumer AI spending.

For most people, that creates room for perhaps one or two recurring AI subscriptions before they start asking harder questions.

Imagine someone paying separately for ChatGPT, Midjourney, Perplexity and a meeting assistant. ChatGPT can already research the web and generate images. Their workplace suite may already summarize meetings. The monthly cleanup becomes obvious even if every product is individually good.

Heavy users behave differently because they can justify a larger stack. A developer may genuinely need Claude Code, ChatGPT and an image tool for different jobs. A marketer may use several generation products every day.

The average payer has a much lower ceiling. Once two subscriptions do roughly the same thing, the weaker habit usually loses.

Get the biggest database of
profitable internet businesses

We mapped 300+ proven digital businesses so you can skip the blind trial and error. For each one, you get the site, the revenue numbers, the distribution strategy, the repeatable patterns, and ideas to recreate the model in a different niche, channel, or angle.

Get the full database →

Are people leaving AI, or just swapping AI subscriptions?

People are mostly swapping, consolidating and upgrading AI subscriptions while overall AI use keeps rising.

Menlo Ventures gives us the clearest consumer view. U.S. adult AI adoption increased only modestly, from 61% to 64%, while estimated global consumer spending jumped from about $12 billion to $40 billion. More than half of AI users say they now pay for at least one product, and 52% say they use AI more than they did a year ago.

That could not happen if cancellations represented a broad retreat from AI.

Business purchasing tells a similar story. Ramp found Anthropic gaining 21.8 percentage points of adoption from January through July while OpenAI slipped slightly. More recently, Ramp still shows Anthropic and OpenAI being used by 80% and 77%, respectively, of companies buying foundational-model software. Many companies use both.

Consumers have become more comfortable switching too. Andreessen Horowitz reported that roughly 20% of weekly ChatGPT web users also used Gemini in the same week. Menlo found the average AI user spreading activity across several general assistants.

This market currently has unusually weak loyalty. A user can move one task from ChatGPT to Claude, another from Midjourney to Gemini and another from Perplexity to ChatGPT without leaving AI for a single day.

For vendors, that feels like churn. For users, it feels like cleaning up the monthly software bill.

Which AI subscriptions look safest from cancellation now?

AI tools look much safer when they become a daily workflow, own valuable context or do something that general assistants still cannot replace well.

Coding is the clearest example today. Cursor, Claude Code and Codex can sit directly inside hours of daily work. Specialized voice products also remain relatively defensible: Andreessen Horowitz notes that ElevenLabs has appeared in every edition of its AI rankings, while Suno remains a major music product. General assistants have made image generation commonplace much faster than they have replicated high-end voice production or full music creation.

Deeply embedded products have another advantage. Notion can connect AI to a user’s workspace. Google can connect Gemini to Gmail, Docs, Search, Photos and Android. ChatGPT and Claude are building connectors that pull more tools and personal context into the assistant.

Every integration raises the inconvenience of switching.

The vulnerable tools sit at the other end: low-frequency products with one generic function, little stored context and several free alternatives.

That is a better predictor of future cancellations than brand popularity. The more replaceable the workflow feels, the more aggressively users will prune the subscription.

Building a digital business?

We have mapped 300+ proven internet businesses. You'll get the full breakdown: revenue, distribution, why it works and how to replicate.

GET THE FULL DATABASE → $49

Which AI tools are people canceling most now?

ChatGPT likely produces the largest number of consumer AI cancellations today, while Midjourney and generic standalone AI tools face the clearest structural risk of being cut.

ChatGPT tops the strongest direct cancellation survey we found, and half of surveyed paid users had left the subscription unused for at least a month. Its huge paid base makes high absolute churn almost inevitable. Current usage, revenue and retention evidence still shows an exceptionally strong underlying product, so ChatGPT Plus looks heavily canceled rather than commercially weak.

Midjourney is more concerning. Its high unused-subscription rate comes alongside a fall from the top ten to number 46 in a major longitudinal AI web ranking. Standalone image generation has lost much of its scarcity now that ChatGPT, Gemini and Canva can generate good images inside broader products.

Generic AI writing products face the same pressure. Users who once needed a separate subscription for AI copy can now get strong writing from the general assistant they already pay for. Jasper and Writesonic have both moved toward deeper marketing workflows for exactly this reason.

Perplexity sits somewhere in the middle. Casual search use is easy to replace, while research-heavy users can still get enough extra value from the product to keep paying.

Claude and Gemini deserve a different reading. Both generate cancellation intent because users can easily switch between assistants, yet their current paid-user and usage trends remain strong. Claude in particular is gaining very quickly among businesses, while Gemini benefits from Google’s enormous distribution.

Coding tools look stronger for now. Developers are switching providers aggressively, but usage and revenue across Claude Code, Cursor, Codex and related products still suggest a growing category.

The clearest answer today is that people are cutting extra AI subscriptions before they cut AI usage. ChatGPT Plus probably sees the most cancellations in raw numbers; Midjourney, generic AI writers, simple research tools and basic meeting notetakers face the strongest pressure from bundling; and products tied to daily work remain much harder to remove.

AI tool or category What we see now Cancellation outlook
ChatGPT Plus Very high cancellation/unused signal alongside massive usage and revenue Lots of cancellations, little evidence of underlying collapse
Midjourney High unused rate plus long-term relative traffic decline Strong structural pressure
Claude Pro Cancellation intent exists, while paid and business adoption keep growing Subscription switching more than abandonment
Gemini paid plans Paid overlap is high while Gemini usage surges across Google Standalone plan can be cut without leaving Gemini
Jasper / Writesonic Generic writing increasingly available elsewhere High pressure unless specialized workflows justify the price
Perplexity Pro General assistants now cover much of AI search Vulnerable among casual users, stronger among heavy researchers
AI meeting notetakers Basic summaries increasingly bundled into productivity suites Growing pressure on simple use cases
AI coding tools Strong current usage and revenue across several vendors Provider churn is high, category demand remains strong
Embedded AI in Notion, Google, Canva and similar suites AI sits inside a broader product people already use Much harder to isolate and cancel

OUR METHODOLOGY

This analysis asks which AI tools people are canceling now, but there is no single dataset that measures product-level cancellation rates across the market. Companies rarely disclose comparable churn figures, and canceling a paid plan does not necessarily mean someone has stopped using the product or AI altogether.

We therefore treated the question as a signal-aggregation problem. We compared direct cancellation intent, unused paid subscriptions, retention, paid-user growth, active usage, spending, switching behavior, product overlap, bundling and how deeply each tool is embedded in a recurring workflow.

Direct evidence received the most weight. Self Financial’s survey of 1,272 U.S. adults provides the clearest product-level data on unused paid AI subscriptions and stated cancellation intent. RevenueCat’s 2026 subscription benchmark, covering more than 115,000 apps and over $16 billion in revenue, provides the broader comparison between AI and non-AI retention, churn, conversion and revenue per payer.

We then tested those cancellation signals against independent usage and spending data. Menlo Ventures’ 2026 consumer AI study provides adoption, spending and multi-product usage data, while Sensor Tower and Andreessen Horowitz provide additional evidence on mobile usage, revenue, web rankings, paid-subscriber trends and category movement.

Business data was kept separate from consumer subscription evidence. Ramp’s spending benchmarks and vendor data were used to understand Anthropic/OpenAI adoption, overlap and switching among companies, not as a direct proxy for Claude Pro or ChatGPT Plus retention.

We also looked at structural replacement risk. A tool becomes easier to cancel when its core function is increasingly available inside another product the customer already pays for. For that part of the analysis, we used first-hand product information from Google Workspace, Microsoft Teams, Zoom, Notion, OpenAI, Midjourney, Jasper and Writesonic.

We did not create an artificial numerical cancellation score. A conclusion became stronger when several independent measures pointed in the same direction. When cancellation intent conflicted with strong usage or paid-user growth, we kept that tension visible rather than forcing the product into a simple winner-or-loser category.

Freshness also mattered. We prioritized 2026 evidence wherever available and used older figures mainly for longitudinal comparisons, such as changes in Midjourney’s relative ranking or the evolution of standalone AI categories.

Key sources include RevenueCat’s State of Subscription Apps, Menlo Ventures’ State of Consumer AI, Self Financial’s subscription survey, Sensor Tower’s State of AI 2026, Sensor Tower’s product-level AI app data, Andreessen Horowitz’s Top 100 Gen AI Consumer Apps, Ramp’s Summer 2026 spending benchmarks, Ramp’s Anthropic vendor data, and Ramp’s OpenAI vendor data.

Additional first-hand sources used to evaluate pricing, positioning and bundling include OpenAI on ChatGPT Go, Anthropic on Claude Code, Midjourney’s pricing documentation, Midjourney’s model history, Jasper’s platform documentation, Writesonic, Google on Gemini Personal Intelligence, Google Workspace, Microsoft Teams recap documentation, Zoom AI note-taking, Notion AI Meeting Notes, and Notion’s broader workspace positioning.

Stop testing random ideas

Start from proof. 300+ profitable internet businesses, mapped, broken down, and ready to copy, in one searchable database.

STEAL WHAT WORKS → $49
Steal What Works

Who wrote this?

STEAL WHAT WORKS TEAM

We study profitable internet businesses, take them apart, and write down what actually works: pricing, distribution, growth, packaging. We turn 300+ proven examples into a database so founders can stop testing random ideas and start from proof. Explore the database →

Back to blog