AI search visibility won't sell your indie game. Here's what it actually does.
Zero-click is real, but AI referrals are tiny and no storefront counts web traffic — Valve says so outright. So what is AI search visibility actually for? Reputation, not wishlists.
AI search visibility is worth some of your attention, just not for the reason most SEO posts give. I don't think it'll bring you wishlists. What it does seem to change is what people find when they go looking for you on purpose.
If you're an indie dev wondering whether you should care about turning up in ChatGPT or Google's AI answers, this is what the evidence actually says: how much traffic AI really sends, what moves your visibility, and what Steam's algorithm does with any of it. That last one is the short answer, and it's nothing.
One definition first, because the term gets thrown around loosely. AI search visibility is how often, and how favourably, your name gets referenced inside AI generated answers across platforms like ChatGPT, Gemini, Perplexity and Google AI Mode. It isn't a rank in a list of blue links. It's whether you get cited or recommended inside the summarised answer itself, and whether what's said about you is right.
- 68% of US Google searches now end without a click. That part of the panic is real and well documented.
- AI chatbots send almost no traffic. Across 81,947 sites, they account for 0.25% of visits. Google sends roughly 210× more.
- Steam doesn't count traffic at all. Valve's own docs say store page traffic is not a visibility factor. Not a small factor. Not one.
- AI search visibility is reputation infrastructure, not an acquisition channel. It shapes what a journalist, a publisher scout or a festival juror sees when they look you up.
- What moves it: brand mentions on trusted third-party sites. Unlinked mentions correlate roughly 3× more strongly with AI visibility than backlinks do.
- What doesn't: schema, llms.txt, word count, E-E-A-T. All four are either untested or actively disproven. I checked each one.
- Track citation rate and accuracy, not rank. There is no ranking position inside an AI answer. What matters is how often you're named and whether the description is right.
The number everyone is quoting
What search engines look like now
SparkToro ran the numbers again this June: 68.01% of US Google searches from January to April 2026 ended without a click. In 2024 the same study put it around 60%. Fewer than one in three searches now sends anyone to an actual website.
That's a panel study, so treat it as a shape rather than a precise measurement. It's a Similarweb clickstream panel, no published sample size, and the mobile/desktop split is an assumption rather than something they measured. SparkToro also quietly restated their own 2024 baseline from 58.5% to 60.45% without explaining why. Worth knowing before you build a slide around it.
AI Overviews and the click drop
The stronger evidence isn't the headline number, it's a randomised experiment. In April 2026 two researchers (Agarwal at ISB, Sen at Carnegie Mellon) ran an actual controlled trial: 1,065 people, a Chrome extension that randomly hid or showed AI Overviews, 68,089 searches. Hiding the AI Overview lifted outbound organic clicks from 0.37 to 0.62 per search.
So AI Overviews cut clicks to websites by 39.8%. Causally, not correlationally. And there was no measurable difference in how satisfied people were, which quietly dismantles Google's "we send higher quality clicks now" defence.
The number nobody is quoting
Fine. So the click is dying. The obvious next move is to go chase visibility inside the AI answers instead, and that's exactly what every vendor in this space will tell you to do.
This is where I want to slow down, because the numbers behind that move are far less impressive than the ones above.
ChatGPT's share of a very small pie
You've probably seen the stat that ChatGPT is about 92% of measurable LLM referral traffic. It's true. Previsible measured 6.77 million LLM sessions across 166 sites and ChatGPT was 92.4% of them.
But 92% of what, exactly. Across 166 properties that's about 3,900 AI sessions per site per month. Previsible themselves note that they excluded Google AI Overviews, which almost certainly represent more AI-driven traffic than every standalone chatbot combined.
Traditional search still dwarfs AI search results
Chartbeat, who have no AI tool to sell you, put it plainly in March 2026: chatbots account for less than 1% of all publisher page view referrals. Meanwhile Google referrals to those same publishers fell 34% year on year, and 60% for small publishers. The traffic isn't moving to AI. It's just gone.
That's the shape of it. Traditional search results are shrinking as a source of visits, and AI search results are not catching what falls. Both things are true, and most of the coverage only tells you the first one.
One more thing worth knowing, because it explains a lot of the noise. In May 2026 OpenAI changed how prominently it surfaces links in answers, and ChatGPT referrals jumped 157% in a single week. In November 2025 a different change halved them overnight. AI referral volume is currently set by product decisions at OpenAI, not by how many people use it. Anyone selling you a growth curve is selling you someone else's roadmap.
No storefront counts your web traffic
This is the part that reframed the whole thing for me. Steam is the only store that says it out loud, so let's start there.
Steamworks has a page called "Visibility on Steam". Two of its section headers are, and I'm quoting the headers verbatim:
Store Page Traffic Doesn't Matter
Conversion Rate Doesn't Matter
The body text is blunter still. "The traffic landing on your store page is not a factor in your Steam store visibility." What Valve counts instead: purchases, playtime, wishlists, follows, tags, and how many languages you ship in.
My first thought was that this is a Steam quirk. It isn't. I went looking for a storefront that treats external traffic as a visibility signal, and I found exactly one.
| Storefront | What drives visibility | Web traffic? |
|---|---|---|
| Steam | Purchases, playtime, wishlists, follows, tags, languages | No, explicitly |
| Nintendo eShop | Revenue, last 3 days. Curated featuring. | No evidence |
| PlayStation Store | Editorial. You submit a form. | No evidence |
| Xbox | Algorithmic plus editorial. Undocumented. | No evidence |
| Epic Games Store | Editorial, or top sellers and revenue. | No |
| App Store / Play | Relevance, downloads, ratings. | Only via downloads |
| itch.io | Page views, weighted to recent. | Yes. The only one. |
A few of those need unpacking.
Nintendo changed its Top 30 charts in April 2025, from downloads over 14 days to revenue over the last 3 days. Cheap games got quietly pushed out of the charts. Nothing in that formula has anything to do with the web.
Apple and Google are the interesting near-miss. Apple has a "Web referrer" bucket in its analytics, so it measures where your web traffic comes from. It has never listed it as a ranking factor. External traffic reaches the algorithm only through the downloads it converts, at the same weight as any other download. Traffic that bounces is worth zero. And on Google Play, the popular claim that backlinks help your ranking has actually been tested with a control group. They don't.
itch.io is the one real exception, and it's narrow. Its founder describes the Popular sort as a mix of historical and recent signals with views as the substrate, and a view is a view whoever sent it. He also says, in the same post, that they delist people who try to game it. And Chris Zukowski's survey of 169 itch developers contains a sentence I'd tape to your monitor: "You will not make money from Itch.io. Nobody does."
What external awareness actually does
So if no algorithm is listening, what is any of this traffic for?
Simon Carless has the best answer I've seen, from years working on the mountain biking game Descenders. The game "sold continually well on console, despite almost no dashboard visibility." Why? "Players were just searching for its name directly."
That's the whole mechanism, and it took me embarrassingly long to see it. External awareness doesn't feed the store's algorithm. It makes someone type your game's name into the store's own search box.
"Best cozy farming games on Steam Deck?"
Assembled live from what the web says. You're named, or you aren't.
No click needed. They just walk away with it.
And the storefront rings it up. It never knew where they came from.
The storefront is a checkout counter, not a discovery engine. Which is finally where AI search visibility connects. Not as a traffic channel. As the thing that decides whether someone walks away from an AI answer knowing your game exists, knowing its name, and knowing it accurately enough to go and type it.
Where store traffic actually comes from
Back on Steam, where we do have numbers. Zukowski surveyed 75 games and found the Discovery Queue alone accounts for somewhere between a third and half of all page visits during launch. The split runs the opposite way to what you'd expect.
For comparison, in Zukowski's client data Twitter accounted for between 0.08% and 1.78% of a game's total traffic, launch months included. That's the bar any new channel has to clear before it's worth your evening.
One more thing about external traffic, and it's the reason content marketing struggles here. Where it counts on Steam, it counts as an ignition signal. Valve watches for spikes and decides within about three days whether to keep showing you. A blog post that trickles in forty visitors a week is structurally the wrong shape for that. It's not a small version of a spike. It's a different thing.
So who actually looks you up?
Right, so if AI search visibility isn't going to sell copies, why am I still recommending you care about it?
Because a specific and disproportionately important set of people will type your studio name, or your game's name, into Google or ChatGPT. Not to discover you. They already know you exist. They're checking you out.
Who checks up on your game
- Journalists deciding whether your press email is worth twenty minutes.
- Publisher scouts doing a first pass before they reply.
- Festival and awards juries vetting a submission.
- Potential collaborators working out if you're real.
- Players who heard your name once from a streamer and are now checking whether the game is any good.
That last group is the one people underrate. Steam's internal search is mostly people typing a game's name, which means the discovery already happened somewhere else and search is just the retrieval step.
That's the actual job. Not ranking. Being legible to a machine that's summarising you to someone who already cares.
What AI systems actually know about your game
It helps to know where the answer comes from, because it's two different things wearing one coat.
Absorbed months or years ago and frozen. If your game shipped after the cutoff, you simply aren't in here.
Mid-answer, the model runs a search, reads a few pages, summarises them. Live, every time. This is the part you can change.
So when someone asks ChatGPT, Perplexity or Google AI Mode about your game, the AI generated answer is mostly assembled on the spot from whatever the web currently says. Your brand's presence in AI search results is really a question about what exists to be retrieved.
If the honest answer is "a Steam page and nothing else", that's roughly what comes back. Sometimes with confident invented details attached, because these systems would rather produce a fluent AI generated response than admit a gap. That's the part worth caring about. Not your position in some list. The accuracy of how your game gets described to someone who is deciding whether to email you.
What actually moves AI search visibility
The good news is that this is much less exotic than the GEO industry wants you to believe. The evidence is unusually clear, partly because some of it comes from vendors publishing findings that undercut their own sales pitch, which I tend to trust more.
| Thing | Does it move AI visibility? |
|---|---|
| Being mentioned elsewhere (linked or not) |
Yes. The strongest single correlate. ρ = 0.664 across 75,000 brands. |
| Backlinks | Weakly. ρ = 0.218. About a third as strong. |
| Schema / structured data | No. −4.6% / +2.4% / +2.2% across three platforms, from 1,885 pages against matched controls. |
| llms.txt | No. 97% of them are never fetched by anything, ever. |
| Being indexed with a snippet | Mandatory. Google says it's a hard gate. |
Brand visibility beats backlinks
The top three factors in that table are all off-site. That's the finding, and it's an awkward one for anyone selling on-page work.
Ahrefs looked at 75,000 brands and ranked what correlates with turning up in AI Overviews. The thing at the top is not the thing the SEO industry sells.
Brand visibility, in the boring literal sense of people saying your name in places you don't control, is the lever.
Notice what isn't on that list: most of what gets sold as content optimization, and nearly all of the technical fixes. The technical bar is genuinely low. Your page has to be indexed and eligible to show a snippet. That's the gate Google names, and it's the only one. Clear it, and the rest of your online visibility comes down to whether anyone is talking about you.
Two of those rows deserve a closer look, because they're the ones you'll get advice about.
llms.txt is decoration. Ahrefs checked all 137,210 domains in their analytics product. About 28% had published an llms.txt. Of those, 97% received zero requests in May 2026. Not from AI crawlers. Not from humans. Nothing. Of the requests that did land on the remaining 3%, the AI retrieval bots that actually determine citations, OAI-SearchBot and PerplexityBot, made up 1.1%. Slackbot fetched llms.txt more often than PerplexityBot did.
Google put it in writing this summer, in their official guidance on generative AI features: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search." The same document says structured data isn't required either, and that "optimizing for generative AI search is optimizing for the search experience, and thus still SEO."
Schema doesn't do what people think. AI-cited pages are about 3× more likely to have JSON-LD, which is the kind of stat that sells a plugin. But Ahrefs tracked 1,885 pages that added schema and compared them against matched controls. AI Overview citations went down 4.6%. The other two platforms moved by amounts indistinguishable from zero. Keep your schema for rich results. It is not the AI unlock.
What I've seen myself, and why you shouldn't lean on it
My day job is in finance, not games. Over the past several months I've been optimising pages there for AI search, doing roughly the things in that table: getting mentioned on other people's sites, writing pages that answer a question plainly, leaving URLs alone once they exist.
Two things shifted. We get more enquiries through the forms, and the people filling them in turn up better informed. They already know what we do. The conversation starts further along than it used to.
Now, every reason not to trust that. It's one company. There's no control group, so I can't separate the AI work from everything else we changed in the same months. And finance is a high-consideration purchase, exactly the sort of thing people research carefully before contacting anyone. Semrush's own data shows citation rates climb most in finance, so it may be the friendliest possible test case.
So I'd call it a thing I observed, not a thing I'd generalise. And notice what improved: not traffic volume. The quality of the people who arrived, and how much they already knew before they arrived. If any of that transfers to games, I'd bet on it transferring to the press-and-publisher side rather than the player side. Which is more or less what the rest of this post argues anyway.
Five things you'll be told to do that don't work
I went looking at what currently ranks for "ai search visibility" and it's almost entirely tool vendors. Here's what they'll tell you, and what the evidence says. I checked each one before writing this, because a couple of them nearly made it into this post.
Semrush's figure, drawn from a single vertical: digital marketing and SEO topics. Their own. No published sample size, no conversion definition. Adobe's dataset of over a trillion retail visits found AI traffic converting 9% worse around the same time. Where it does convert well, that's because those people finished their research in the chat window. Selection, not a lever you can pull.
I couldn't trace this to an actual study, and it's correlation regardless. Ahrefs added schema to 1,885 pages and compared them against matched controls: AI Overview citations went down 4.6%. Google's own words: "Structured data isn't required for generative AI search."
E-E-A-T is a concept from Google's guidelines for its human quality raters. Google says plainly that rater data "is not used directly in our ranking algorithms" and that E-E-A-T "isn't a specific ranking factor." Google's own guidance on AI features never mentions it once. No LLM has an E-E-A-T module. It's an SEO framing retrofitted onto a mechanism it has nothing to do with.
A Semrush correlation study from 2019, with the baseline it was measured against (901–1,200 words) quietly dropped somewhere in transit. Google: "There's no ideal page length." John Mueller: "Word count is not a ranking factor. Save yourself the trouble."
Across 17 million citations, AI-cited pages are about 25% newer than organic results. But Google's AI Overviews cite older pages than organic search does, and the average AI-cited article is still nearly three years old. Google lists artificial freshening as a warning sign you're doing it wrong.
Generative engine optimization is still SEO
Notice the pattern. Every one of those claims is either a correlation dressed up as a mechanism, or a stat that lost its baseline somewhere in transit. That's not a coincidence. Generative engine optimization is being sold as a new discipline, and a new discipline needs new rules to sell.
Google's position is refreshingly blunt: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."
Where AI SEO and traditional SEO actually differ
That said, I don't want to flatten this into "nothing has changed", because two things have.
The first is that there's no position to win. Traditional SEO hands you a rank you can track over time. AI hands you a probability. You either get named often, or you don't, and the ordering inside the answer is noise.
The second is the weighting. In classic SEO a mention without a link is worth little. In AI SEO it appears to be worth more than the link. That's the one genuine shift in emphasis I'd act on. Everything else on the checklist is the same checklist it always was.
What about AI search visibility tools?
There are more every month. An AI search visibility tool tends to promise the same handful of things: a visibility score, prompt tracking across multiple AI platforms, some competitive intelligence on who else gets named alongside you, and a share of voice figure for your brand. A few of the traditional SEO tools you already pay for have bolted this on. Others are standalone AI visibility checkers.
Platform coverage: ChatGPT, Gemini, Perplexity and Google AI Mode
Coverage is the first thing I'd look at, and it varies wildly. Most tools handle ChatGPT, Gemini and Perplexity. Fewer do Google AI Mode properly. Almost none touch Meta AI.
If a tool reports on one platform only, treat its numbers with suspicion, because different AI engines cite almost different webs. One study found only about 11% of domains are cited by both ChatGPT and Perplexity. A score built on a single platform isn't a view of your AI visibility. It's a view of that platform's habits.
What a visibility score actually tells you
Some of what these tools sell is useful. One part of it is not, and it's worth knowing which is which.
SparkToro and Gumshoe ran 2,961 tests across ChatGPT, Claude and Google's AI surfaces. They found under a 1 in 100 chance that two runs of the same prompt return the same list of brands. Under 1 in 1,000 that they come back in the same order. Every run varies in what it names, in what sequence, and in how many things it names at all.
So any tool selling you a ranking position inside an AI answer is selling you noise. There is no position. There's a dice roll.
Checking how your brand performs, for free
You don't need to buy anything to do this, and for a one-person studio I'd skip the tools entirely for now.
Write down ten industry relevant prompts. Not keywords, actual sentences. The target queries a real player or a real journalist would type. "Best cozy farming games on Steam Deck." "Is [your game] any good." "Games like [the obvious comparison everyone makes]." Then run each one twenty times, in a fresh chat every time, and count how often your game gets named.
That's your visibility data. It's tedious and it takes an evening, but it's the same key metric the paid tools sell you, measured the same way. And while you're in there, read what the AI actually says about your game. Half the time the interesting finding isn't whether you appear. It's that the description is wrong.
What I'd actually do
Given all that, here's the order I'd put things in. It's deliberately boring.
Capsule, first five seconds of the trailer, tags, short description. And on console especially: the words in your title, since in-store name search is the channel that responds to you. This is the conversion surface for every visitor you'll ever get.
Unlinked mentions correlate roughly 3× more strongly with AI visibility than backlinks. A Reddit thread, a podcast, a devlog someone else writes about you. That's the lever.
Plain HTML, indexed, snippet-eligible, honest about genre and comparisons. That's the page an LLM will summarise when a journalist asks about you. Keep the URL stable forever.
Genuinely. The evidence says it's decoration. Spend that afternoon on your capsule art instead.
Streamers, a festival, a Reddit post that lands. Valve's algorithm watches for bursts. Content marketing produces a drip, which is the wrong shape.
None of this is exciting, and that's sort of the point. Around 20,000 games shipped on Steam in 2025 and roughly 600 of them cleared a thousand reviews. The gap between those two numbers is not an SEO gap.
Honestly, I think a lot of indie "SEO strategy" is displacement activity. It's legible, it's controllable, you can do it on a Tuesday evening and feel productive. The actual job is harder: make a game with a hook strong enough that a stranger scrolling a storefront stops. Every store in that table is, in its own way, built to be immune to everything except that.
So yeah. AI search visibility is worth some attention, as long as you're clear-eyed about what it's for. It's the thing that makes you look real to the people who come looking. I really don't think it's going to sell your game.
Sources
Search and AI traffic
- SparkToro — In 2026, less than one third of Google searches still send a click (Jun 2026)
- Agarwal (ISB) & Sen (CMU) — the first randomised experiment on AI Overviews (Apr 2026, revised Jun 2026)
- Ahrefs — AI traffic across 81,947 sites
- Previsible — 2026 State of AI Discovery Report (Jul 2026)
- Chartbeat, via Axios — chatbots send under 1% of publisher referrals (Mar 2026)
- Similarweb — ChatGPT referral traffic triples after an OpenAI product change (May 2026)
What moves AI visibility
- Google — Optimizing your website for generative AI features on Search
- Google — Creating helpful content (on word count and artificial freshness)
- Ahrefs — AI Overview brand visibility factors (75,000 brands)
- Ahrefs — We tracked 1,885 pages adding schema (May 2026)
- Ahrefs — 97% of llms.txt files never get read (137,210 domains, Jun 2026)
- Ahrefs — AI assistants prefer to cite fresher content (17M citations, Jul 2025)
- Semrush — The impact of AI search on SEO traffic (Jul 2025), the source of the 4.4× claim
- Adobe Analytics — Generative AI traffic to US retail sites (Mar 2025)
- John Mueller — "Word count is not a ranking factor"
- SparkToro & Gumshoe — AIs are highly inconsistent when recommending brands (2,961 runs)
- Profound — AI platform citation patterns (only ~11% domain overlap between ChatGPT and Perplexity)
Storefronts and game discovery
- Valve — Visibility on Steam (Steamworks documentation)
- Chris Zukowski — What is the post launch Discovery Queue (75 games, Feb 2025)
- Chris Zukowski — Where visibility comes from
- Chris Zukowski — itch.io traffic benchmarks (169 developers, May 2025)
- Simon Carless, GameDiscoverCo — What a banned PlayStation dev tells us (the Descenders story, Mar 2026)
- Simon Carless, GameDiscoverCo — The biggest discovery change in Switch history (eShop charts move to revenue, May 2025)
- Sony Interactive Entertainment — Showing your game to PlayStation (Apr 2026)
- Xbox Wire — ID@Xbox at GDC 2026 (on algorithmic promotions)
- Epic Games Store — Marketing documentation (how store blades are ranked)
- leafo (itch.io founder) — How itch.io sorts games
- Apple — App Store search and App Store Connect acquisition sources
- Google Play — How apps are ranked and discovered
- yellowHEAD — Do backlinks impact Google Play ASO? (a controlled experiment: no)
- SteamDB — Steam release statistics
