Does Adding Original Screenshots Change AI Search Visibility? A 918Kiss Experiment

The Question That Started With a Picture

Scroll through enough 918Kiss guides and the pictures start to blur together. The same glossy slot reels, the same generic phone mock-ups, the same images that could sit on any page about any app. For a long time that did not seem to matter. A page ranked, a reader clicked, and the picture was decoration.

AI search changed the stakes. When Google answers a question with an AI Overview or in AI Mode, when Microsoft Copilot writes a summary, or when ChatGPT search pulls sources into a reply, only a handful of pages get named. Everything else stays out of sight, however well it once ranked. So a sharper question started to bother us: if a page shows real, original screenshots of the 918Kiss app, taken by a person on a real phone and showing the exact screen the text describes, is that page more likely to be seen and cited by AI search?

This is the story of how we set out to answer that honestly. It is not a results announcement. It is the experiment itself: what the official documentation says, where it goes quiet, and the test design that lets data, not a hunch, deliver the answer.

What We Hoped Was True

The hope was simple and a little romantic. AI systems are trying to find trustworthy sources. A real screenshot looks like proof that someone actually opened the app, tapped through the menus and saw what they are describing. Surely that kind of first-hand evidence would stand out among pages built on stock images.

It is a good story. But a story is not a finding, and the fastest way to waste months of work is to believe something because it feels right. Before touching a single page, we went to the sources that actually decide how search works.

What Google Says, and Where It Stops

Google’s own page on AI features is surprisingly plain. It says there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary,” and that there is “no special schema.org structured data that you need to add.” To be shown, a page must be indexed and eligible to appear in Google Search with a snippet. That is all.

Among the ordinary best practices the same page lists, one line matters for this experiment: “Supporting your textual content with high-quality images and videos, when applicable.” Another says your structured data should match the visible text on the page. The page also explains that AI Overviews and AI Mode may use a “query fan-out” technique, running several related searches across subtopics to find a wider set of helpful links.

Google’s image SEO guide fills in how images are understood. Google extracts information about an image “from the content of the page, including captions and image titles,” and asks site owners to place images “near relevant text.” It calls alt text “the most important attribute” for image metadata and says Google uses it “along with computer vision algorithms and the contents of the page.” Descriptive file names give “very light clues.” Images should sit in a normal <img src>element, because Google doesn’t index CSS background images, and should use a supported format such as JPEG, PNG, WebP or AVIF.

Then the documentation stops. Nowhere does Google say an original screenshot is a signal for AI Overviews, AI Mode or anything else. It says high-quality images support good content. It does not say they earn citations. That silence is the gap this experiment is built to fill.

The Turning Point: Finally Being Able to See

For a long time, a test like this would have been mostly guesswork, because there was no clean way to see AI visibility. Google’s AI features page says appearances in AI Overviews and AI Mode are counted inside the normal Search Console Performance report, under the “Web” search type, mixed in with everything else.

2026 changed that. In June, Google introduced Search Generative AI performance reports in Search Console. According to coverage of the launch, they show impressions in AI Overviews and AI Mode broken down by page, country, device and date, though not clicks or queries. In late September, Google added multimodal reporting. Industry write-ups describe a “Web: multimodal” view covering searches made with Google Lens, Circle to Search, image uploads and Chrome’s right-click image search. Microsoft had already added an AI Performance report to Bing Webmaster Tools in February 2026, showing total citations, average cited pages, the grounding queries the AI used to find a page, and which URLs get cited most.

That was the moment the experiment became real. But it brought a second realisation with it. Taking a real screenshot forces the writer to look at the real screen, and that usually changes the words too. The caption gets more specific. The paragraph next to the image starts naming the actual buttons, menus and steps. So if a page with screenshots does better, is it the picture or the sharper text the picture inspired? A fair experiment has to separate the two.

Designing the 918Kiss Experiment

Choosing the pages

We start with twelve existing 918Kiss guide pages on similar topics, such as downloading, logging in, account settings and game rules, with roughly similar traffic. They are split into three groups of four:

• Control: nothing changes.

• Group A, screenshots only: existing images are replaced with original screenshots, with new alt text and captions. The body text stays the same.

• Group B, screenshots plus walkthrough: the same screenshot treatment, plus a short paragraph beside each image describing exactly what the screen shows and what to tap next.

If Group B pulls ahead and Group A does not, the words are doing the work. If Group A also moves, the images themselves may matter.

Taking screenshots that deserve to exist

Every screenshot is captured on our own device, on the current app version, and cropped to the part of the screen the text talks about. Usernames, user IDs, phone numbers, balances and transaction references are blurred before upload. Nothing is edited to make a screen look better than it is. That means no added wins and no doctored balances. A screenshot that misleads a reader is worse than no image at all.

This is also where responsible play lives naturally in the content. The most useful screens to document are often the quiet ones: where to find account settings, how to read a game’s rules, where transaction history sits. A guide that shows readers those screens is helping them stay in control, not chasing a jackpot image.

Each file gets a descriptive name, such as 918kiss-android-install-settings.webp, alt text that says plainly what is on screen, and a visible caption. Each image sits directly beside the paragraph it illustrates and is compressed so it does not slow the page down.

Measuring what changes

The test runs on a fixed calendar. There are 28 days of baseline before the change, then all changes go live on the same day, then 56 days of observation. For every page we record:

• Impressions in the Search Generative AI performance report

• Activity under the multimodal view in Search Console

• Standard clicks and impressions in the Performance report

• Citations, cited pages and grounding queries in Bing Webmaster Tools’ AI Performance report

• A weekly manual log of the same set of prompts in Google AI Mode, Copilot and ChatGPT search, noting which of our URLs, if any, are cited

For ChatGPT, one technical check comes first. OpenAI says sites that opt out of its OAI-SearchBot crawler are not shown in ChatGPT search answers, so the robots.txt file must not block it.

Deciding what counts as an answer

The comparison is always against the control group, not just before against after. Google updates, seasonal swings and news cycles can move every page at once, and a control group is what keeps those from being mistaken for an effect. With four pages per group, a small difference is not proof. We will report what moved, how much and in which group, and say clearly when a result is too small to mean anything.

Repeating the Test on Mega888 Pages

One result on one set of pages is a clue, not a rule. So once the 918Kiss run finishes, the same design is repeated on Mega888 guide pages, with the same three groups, the same screenshot rules and the same measurement calendar. If the pattern holds on a second, separate app topic, it starts to look like something real. If it only appears once, that is worth knowing too.

Why Start Before the Answer Arrives

It would be easy to wait for someone else to publish the answer. But AI search is new enough that nobody outside the search companies can say with certainty what earns a citation. The only reliable way to know what works on your own pages is to test on your own pages.

And whatever the data says, the readers win. A guide with real, honest screenshots of the actual screens is clearer than one decorated with stock art. If AI search rewards that, wonderful. If it doesn’t, the page is still better for the people who matter most. The results will follow here when the experiment is complete.

Frequently Asked Questions

Does Google require special markup for images to appear in AI Overviews?

No. Google says there are no additional requirements or special optimizations for AI Overviews or AI Mode, and no special schema.org markup is needed. A page has to be indexed and eligible for a snippet. Normal image best practices, such as alt text, captions and placement near relevant text, still apply.

Can Search Console show whether a page appeared in AI Overviews?

Yes, since 2026. The Search Generative AI performance reports show impressions in AI Overviews and AI Mode by page. Coverage of the launch notes that they do not show clicks or query data.

If AI can see images, does alt text still matter?

Yes. Google describes alt text as the most important image attribute and says it uses alt text along with computer vision and the page’s content to understand what an image shows.

Is it safe to show my own account in screenshots?

Only after blurring anything personal, including usernames, IDs, phone numbers, balances and transaction references. Never edit a screenshot to show results that did not happen.