What the research says about product screenshots

A screenshot should show the product as it is today. That is Screenery's whole premise, so this page collects the evidence behind it and says how strong each piece is. Every number links to the page it came from. We re-opened each source in November 2026 before citing it, and dropped the ones we could not open.

The evidence

Finding	Number	Source	Evidence	Year
SaaS sales sites that don't show their product's UI clearly enough	35% of the sites in Baymard's SaaS benchmark	Baymard Institute	Independent research: usability testing and a benchmark. No sample size published.	2022
Which media prospects use on a SaaS site	Images rank first. GIFs did as well as images in testing. Video was used by only a handful of participants, and demos by few	Baymard Institute	Independent research: usability testing. No sample size published.	2022
B2B buyers who prefer an overall rep-free buying experience	61% of 632 buyers, surveyed August to September 2024	Gartner	Analyst survey	2025
B2B buyers who saw inconsistencies between a sales organization's website and what its sellers told them	69%, same survey	Gartner	Analyst survey	2025
B2B buyers who prefer a rep-free experience	67% of 646 buyers, surveyed August to September 2025. A different sample and wording from the 2025 release	Gartner	Analyst survey	2026
B2B buyers who used AI during a recent purchase	45%, same survey	Gartner	Analyst survey	2026
How far through a purchase buyers are before they engage sellers	"Nearly 70%". Buyers make first contact more than 80% of the time	6sense	Vendor survey. Sample size not in the release.	2024
Split between independent research and talking to sellers	Moved from 70/30 to 60/40, in a survey of more than 4,000 buyers. 94% of buying groups ranked preferred vendors before first contact	6sense	Vendor survey	2025
When B2B software buyers find it most useful for sales to contact them	Research stage, 34%. Only 3% never want to talk with sales. More than 1,900 buyers	G2	Vendor survey	2024
What B2B sites should show	Show the product in action, offer as much information as you can without contact with sales, and show the price	Nielsen Norman Group	Expert guidance	2019
Task success on B2B sites	58%, against 66% on mainstream sites. 79 participants, 179 sites. Missing prices were the most user-hostile element	Nielsen Norman Group	Independent research: qualitative usability study	2006
Viewing time on product photos on one category page	18% on photos, 82% on text. People look at images that carry information and ignore decorative ones	Nielsen Norman Group	Independent eyetracking research. Sample size not stated; product photos, not software	2010, reviewed 2026
Credibility comments that mentioned a site's "design look"	46.1% of comments, from 2,684 people. A share of comments, not of people	Consumer WebWatch	Independent research	2002
Open-source survey respondents who had seen incomplete or outdated documentation	93%. A perception, not a measurement, and incomplete and outdated are not separated	GitHub Open Source Survey	Platform survey: 5,500 randomly sampled respondents	2017
Top obstacle to consuming APIs	Lack of documentation, 54.3%. More than 13,500 respondents	Postman State of the API	Vendor survey	2020
VS Code issue reports about outdated screenshots	210. 455 of VS Code's 590 docs pages have screenshots	Tang et al.	Peer-reviewed	2026
Screenshots a 297-day age rule flags as outdated, against those that were	The rule flagged 70.3%. Of 154 flagged screenshots checked by hand, 8 were outdated	Tang et al.	Peer-reviewed	2026
Outdated screenshots a detector found in VS Code's docs	17. VS Code confirmed and fixed every report. It also found 20 across 50 Android app websites	Tang et al.	Peer-reviewed	2026
Requests to Mintlify-hosted docs from AI coding agents	45.3% of about 790 million requests over 30 days, counted by user agent (March 2026)	Mintlify	Vendor data, from its own customers' sites	2026
Agents' share of measured traffic to Mintlify-hosted docs	66% in July 2026, a partial month, counted a different way	Mintlify	Vendor data, from its own customers' sites	2026
Learning from words and pictures, against words alone	Median effect size d = 1.35 across 13 experimental comparisons. Short lab lessons, not software docs	Richard E. Mayer	Peer-reviewed research program, summarized by its author	2023
Full-screen and partial-screen captures in software manuals	Text-only and full-screen manuals both beat partial-screen captures for learning. No difference when used as a job aid	Gellevij et al.	Peer-reviewed. Sample sizes not stated in the abstracts	1998, 1999
A software manual with purpose-built screenshots, against text only	Training 11% faster and 14% higher scores on trained tasks with screenshots. 44 students	Gellevij et al.	Peer-reviewed, one study	2002
Screenshots on an App Store product page	Up to 10. Depending on orientation, up to three can appear in search results	Apple	Platform guidance	Current
Text alternatives for images	Non-text content needs a text alternative that serves the equivalent purpose: success criterion 1.1.1, Level A	W3C, WCAG 2.2	Standard	Current

How strong is it?

Each row carries a label. They are not equal, so read the label before you repeat the number.

Peer-reviewed: published after review by other researchers. The strongest kind here.
Independent research: usability testing or a study by a research firm that sells no product the result would help.
Analyst survey: a survey by an analyst firm such as Gartner, with the sample size and dates published.
Vendor survey or vendor data: a company reporting on its own survey or its own customers. Useful, but it has a stake in the answer, and the method is often thin.
Platform guidance and standards: what Apple or the W3C tells you to do. Rules, not measurements.

Buyers

B2B buyers do most of their research before they talk to anyone. In a Gartner survey of 632 buyers, 61% preferred an overall rep-free buying experience, and in the same survey 69% reported inconsistencies between a sales organization's website and what its sellers told them. A screenshot that no longer matches the product is that kind of inconsistency. What they look for first is the product itself: Baymard found 35% of SaaS sales sites don't show the UI clearly enough, and of the media it tested, images mattered most.

What B2B buyers look at before they talk to you has the rest, including the 6sense and G2 surveys and why the two Gartner numbers are not a trend.

Docs readers

Outdated screenshots are common and they confuse people. A 2026 study of VS Code's documentation found 210 issue reports about outdated screenshots, some from developers with more than ten years' experience. Its most useful finding is that age is a poor test. An age rule flagged 70.3% of VS Code's screenshots as outdated, and only 8 of 154 flagged ones checked by hand were.

Pictures help learning in general: Mayer's lab studies found people learn more from words and pictures than from words alone. Studies of real software manuals are more mixed. One found full-screen captures and text-only manuals both better for learning than partial-screen captures. A later one, where each screenshot had a job, found people trained 11% faster and learned more.

Read How outdated screenshots show up in real docs and Do screenshots in instructions help?

AI agents

On one vendor's numbers, coding agents now request docs about as often as browsers do. Mintlify counted 45.3% of requests to the docs it hosts coming from AI coding agents over 30 days in March 2026, and browsers 45.8%. A later report with a different method put agents at 66% of measured traffic in July 2026. An agent learns your UI from the text around a screenshot, its alt text, and sometimes the image. If those are stale, it teaches the old screen.

Your docs' newest readers are AI agents covers both methods and what to change.

What doesn't hold up

Four statistics about screenshots and visuals get repeated far more than they get checked: "60,000 times faster", "80% of knowledge bases are out of date", "the first three app-store screenshots drive most installs", and "7 seconds on a store listing". None of them holds up the way it is usually quoted. Screenshot statistics that don't hold up traces each one and says what to cite instead.

What nobody has measured

Two numbers would settle the business case, and we could not find either.

How much a stale screenshot lowers conversion or raises support tickets. No source we found measures it, so we don't estimate it.
The share of buyers who rely on screenshots specifically, as opposed to demos, reviews or trials.

If you know of a study that measures either, we would like to read it.

What this means for your screenshots

Show the real product, prominently. Prospects look for the UI, and 35% of SaaS sales sites under-show it (Baymard). NN/g's advice is to show the product in action.
Make every screenshot match the product. Buyers already notice when a site and a seller disagree (Gartner, 69%), and readers file bugs when a screenshot doesn't match the screen (VS Code, 210 reports).
Check screenshots against the app, not the calendar. An age rule flagged 70.3% of VS Code's screenshots, and almost all of them were fine (Tang et al.).
Show enough of the screen to find the control. Partial-screen captures did worse for learning than full-screen ones in one study (Gellevij et al., 1999). Crop to what matters, but keep enough around it to orient the reader.
Put the picture beside the step, and describe it in alt text. Words and pictures together teach better than words alone (Mayer), and every image needs a text alternative (WCAG 1.1.1, Level A).
Make the page work without the image. Agents now make a large share of docs requests (Mintlify), and whether one looks at an image depends on the tool. The steps should make sense from the text and alt text alone.

Screenery handles rules 2 and 3 for you. Your tests take the screenshots, the default branch publishes them at a URL that always serves the newest build, and a pull request shows which screenshots changed against @latest. See How to keep documentation screenshots up to date.

Sources

Every page below was opened on 13 November 2026, and each number on this page was checked against it.