• AI visibility

Guide · AI search

How do AI image generators get recommended by AI assistants?

By being specific about what they do best and clear about commercial rights, then getting that story into the independent reviews, galleries, tutorials and “best of” lists that AI assistants read. Image generation is crowded, the leading chat assistants now make images themselves, and business buyers ask about licensing before quality. A generator that answers the rights question plainly gives assistants, and buyers, a reason to name it.

By Underneath Team Guide Updated

The short version

  1. ChatGPT is now an image tool too: multimedia requests grew from 2% to just over 7% of its messages, with a spike after its new image generation launched in 2025, per a study by OpenAI and Harvard economists.
  2. Paid growth is expensive in this category: top-quartile Photo & Video apps pay above $14 per install on iOS, per RevenueCat’s (opens in a new tab) 2025 benchmarks.
  3. Business demand is real: 53% of B2B marketers whose organizations use AI use creative asset tools for images and video, per the Content Marketing Institute (opens in a new tab).
  4. Rights terms differ sharply: Midjourney (opens in a new tab) requires companies with more than $1,000,000 a year in revenue to be on a Pro or Mega plan to own their images, while Getty Images (opens in a new tab) offers uncapped indemnification.
  5. Rules are tightening: the US Copyright Office concluded that prompts alone do not make users authors, and the EU AI Act’s Article 50 (opens in a new tab), which comes into force on 2 August 2026, requires machine-readable marking of generated images.

Who uses AI image generators, and what is a customer worth?

Individual creators on monthly plans, plus marketing teams and brands that pay more for rights and control.

There are three broad buyers. Individual creators and hobbyists want quality and style. Marketers want speed: in the Content Marketing Institute’s survey, creative asset tools were the second most used kind of AI tool, at 53% of marketers whose organizations use AI. Brands, agencies and media companies want images they can use without legal risk, and they will pay for that.

Pricing reflects those tiers. Midjourney’s plans page (opens in a new tab) lists four subscriptions from $10 to $120 a month, with a 20% discount for paying a year upfront. Its terms tie ownership of images to plan level for larger companies, which pushes business users to the higher tiers. Stock providers sell rights as the product: Getty says customers get “representations and warranties, uncapped indemnification” on images from its generator.

The consumer economics are attractive but uneven. For AI apps in general, RevenueCat put typical earnings above $0.63 per install at the 60-day mark, against a $0.31 median for all apps. In Photo & Video apps, top apps earn 5–7 times more than the median, and 27.57% of apps reached $1,000 in revenue within two years, the highest success rate of any category. The catch is acquisition cost: RevenueCat attributes Photo & Video’s high cost per install to “the competitive landscape.” A user who arrives because an AI assistant named you costs nothing per install, which is why the channel matters here.

What happens to image generators now that chat assistants make pictures too?

They are now image generators themselves, and they are also where users ask which tool to use.

The OpenAI and Harvard study of ChatGPT messages found multimedia requests grew from 2% to just over 7% of usage, “with a large spike in April 2025 after ChatGPT released new image-generation capabilities,” and the level stayed elevated. OpenAI’s announcement (opens in a new tab) of that feature also says all generated images “come with C2PA metadata” identifying them as AI-made. For a casual user, the assistant may be all the image generator they need. Writing tools face the same squeeze; see how AI writing tools still win users.

Specialists still hold ground. In Andreessen Horowitz’s fifth Top 100 list (opens in a new tab) of consumer AI products, Midjourney and Leonardo are among the fourteen “All Stars” that have appeared in all five editions of its web list, and Midjourney is “famously bootstrapped.” We infer that the durable positions are style and quality leadership, control (editing, consistency, brand styles), workflow integration and commercial safety, not simple text-to-image.

Which questions do image-generator users ask AI assistants?

Style, use-case, comparison, price and rights questions, with rights dominating for business users.

We made up the prompts below to show how creators and marketers tend to phrase image-tool requests; they are not logged queries:

  • Style: “Best AI image generator for photorealistic product shots on white backgrounds.”
  • Use case: “AI tool that keeps the same character across a 20-page children’s book.”
  • Comparison: “Midjourney vs Adobe Firefly vs ChatGPT images for a marketing team.”
  • Price: “Cheapest AI image generator with no watermark and unlimited images.”
  • Rights: “Which AI image generators are safe for commercial use and offer indemnification?”
  • Ownership: “Can I copyright images I make with an AI image generator?”

Rights questions are where a wrong or vague answer costs the most. They also have documented answers: the Copyright Office concludes that “prompts alone do not provide sufficient human control to make users of an AI system the authors of the output,” while creative human arrangements or modifications of the output can be protected. A vendor that explains this plainly, alongside its own license terms, gives an assistant something accurate to repeat.

Why do licensing and provenance decide trust in this category?

Because business buyers carry the legal risk, and lawsuits and new rules have made that risk visible.

Litigation is active. NPR reports (opens in a new tab) that Disney and Universal filed a 110-page lawsuit against Midjourney alleging that it trained on “countless” copyrighted works, and that Getty Images has sued Stability AI. The Copyright Office report drew on more than 10,000 public comments, a sign of how contested the topic is. For a marketing or legal team, the question “is this safe to use?” comes before “is this the best image?”

Vendors answer it in different ways, and the differences are concrete:

Rights questionExample of a documented answer
Who owns the output?Midjourney: users own assets, but companies over $1,000,000 in revenue must be on Pro or Mega
Is it protected if someone sues?Getty: uncapped indemnification, trained solely on its own library
Can others tell it is AI-made?OpenAI: C2PA metadata on all generated images
Is it copyrightable?Copyright Office: not on the basis of prompts alone

Provenance is becoming a legal requirement as well as a trust signal. Article 50 of the EU AI Act requires providers of systems that generate images to ensure outputs “are marked in a machine-readable format and detectable as artificially generated or manipulated.” Google DeepMind describes SynthID (opens in a new tab) as “a tool to watermark and identify content generated through AI.” We infer that a generator with clear, public answers on ownership, indemnification, training data and marking is easier for both buyers and assistants to recommend for business use.

How does an AI recommendation turn into revenue for an image generator?

Through a free or low-cost first session that becomes a subscription, and for businesses, a plan bought for rights.

We infer two paths:

  1. Creators. A user asks an assistant for a tool that does a specific style or task, tries one or two named tools, and subscribes if the first results are good. RevenueCat’s data shows how much rides on that first session: across categories, most trial-to-paid conversions happen immediately.
  2. Businesses. A marketer or designer asks which tools are safe for commercial use, shortlists those with clear terms, and the company buys a team or enterprise plan, sometimes with indemnification or an API. Here the assistant’s description of your licensing can win or lose the deal before anyone visits your site.

Neither path shows up neatly in analytics. A creator who heard of you in ChatGPT often arrives by typing your name. For how to measure it anyway, see why analytics miss AI visibility.

What decides which image generators an AI assistant names?

The platforms say little, but citation studies point to ranked lists, independent reviews and recently published pages.

Documented by the platforms. A request like “best image generator for logos” is not searched as typed: OpenAI explains (opens in a new tab) that ChatGPT search typically rewrites it into “one or more targeted queries,” and Google reports (opens in a new tab) that AI Mode fans out into “multiple related searches” across subtopics. Neither company explains how a particular generator ends up in the answer.

Observed in studies. “Best AI image generator” lists are central to this category, and ranked lists are what assistants cite most: in Kumar’s (opens in a new tab) data from an AI visibility platform, the “best-of” listicle was the most-cited content format, at about 21% of all citations. Many such lists are written by vendors. In our study of numbered lists cited by AI, 24.2% of those with an identifiable publisher ranked their own publisher first. Outside coverage was the strongest signal in our brand entity study, where every tenfold increase in independent sites naming a brand came with 4.7 times the odds of being recommended. And assistants that search favor fresh pages: in our freshness study, 17.4% to 22.6% of their dated citations were under 90 days old, against 6.9% of Google’s top 10, which suits a category where models change every few months.

Our inference for image generators. Assistants answer in text, so a generator’s visual quality reaches them only through what others write: reviews with side-by-side tests, tutorials, creator community threads and comparison articles. Rights and pricing reach them through your own pages and how consistently others repeat them. When those disagree, answers can go wrong; see how to fix wrong brand information in AI answers.

What does GEO involve for an image generation product?

It gets your best styles and your licensing terms stated clearly, repeated consistently and backed by outside reviewers. No one can guarantee an assistant will name you.

In practice, generative engine optimization (GEO) for an image tool comes down to six tasks:

  1. A clear entity. Say what you are best at (styles, consistency, editing, product shots, illustration), who you are for and where you work (web, app, API, plugins), the same way on your site, app stores, review profiles and social accounts.
  2. A plain-language rights page. Cover ownership, commercial use by plan, indemnification, training data, content marking and what users may not generate. Keep it current and dated, because these are the questions business buyers ask.
  3. Independent tests. Earn side-by-side reviews from design publications, creators and testing sites, rather than relying on your own “best of” page. Why ranked lists drive AI recommendations covers how those lists get used.
  4. Text that describes images. Publish galleries and use-case pages with written explanations of style, prompts and settings, so text-based answers have something to quote.
  5. Creator community presence. Tutorials, community challenges and genuine participation in the forums where creators compare tools.
  6. Measurement. Run style, use-case, comparison, price and rights prompts through ChatGPT, Gemini, Perplexity, Claude, Copilot and Google at regular intervals, and ask each new subscriber how they found you.

For challengers facing bigger names, see do AI assistants favor big brands.

What is still unknown about AI search and image generator growth?

We have found no study of how often assistants recommend a specific image generator, or how those users convert.

The figures on image requests inside ChatGPT were produced by researchers at OpenAI, working with Harvard economists. RevenueCat’s benchmarks cover mobile Photo & Video and AI apps broadly, not image generators alone. Individual artists and hobbyists are outside the Content Marketing Institute’s sample, which is B2B marketers only. The legal picture is unsettled: lawsuits are ongoing, the Copyright Office left training data and liability to a separate part of its report, and EU marking rules are only now taking effect. Studies of AI citations cover many categories, and how they apply to image generators is our inference. We have also seen no study of how often assistants suggest their own image tools instead of a specialist.

How can an image generator tell whether assistants are costing it subscribers and business plans?

Test whether assistants name you for your strongest styles, and whether they get your licensing terms right.

Put your style, use-case, comparison, price and commercial-use prompts to each major assistant and compare what comes back with your subscription and team-plan numbers; that shows whether the gap is with creators, business buyers or both. For help getting your rights terms and independent reviews into those answers, talk to us about an image generator audit. You can see how we scope that work, including the rights page, independent tests and recurring prompt checks, on our generative engine optimization service page.

Frequently asked questions

Do AI assistants recommend their own image tools over specialists?

That has not, to our knowledge, been measured. Asked for the best tool in the category, assistants do list outside generators, though no platform documents how the selection is made.

Can images from AI generators be copyrighted?

In the US, the Copyright Office concluded that prompts alone do not make a user the author, while creative human arrangements or modifications of AI output can be protected. This is general information, not legal advice.

What makes an image generator “commercially safe”?

There is no single standard. Vendors use the term for different things: licensed training data, ownership terms, indemnification or content marking. Buyers should read the terms, and vendors should state them plainly.

Should image generators publish comparisons with Midjourney or ChatGPT?

Yes, as long as the comparison is fair. Users ask for exactly these match-ups, and an honest page with real side-by-side examples and clear differences gives assistants something specific to repeat.

Sources

Free strategy call

Some questions are easier to answer about your own business.

Bring the one that matters most. On a free 30-minute call we’ll take a first look at it and send you a short written read afterward.

Ask about my businessHow we work