Two apps walked into a bar. Notion drank their lunch.
We ran the Alfred vs Raycast matchup through our /vs tool this week, expecting a real fight. Alfred is the veteran macOS launcher with a cult following that predates the App Store's dominance. Raycast is the slick, extensible, VC-backed challenger that ate a big chunk of Alfred's mindshare over the last three years. If you asked any Mac power user which of the two they'd pick, you'd get a spirited defense of one or the other. Sometimes both, depending on the workflow.
We asked ChatGPT, Gemini, and Perplexity the same question in five different buyer-shaped ways.
Neither brand got mentioned. Not once. In any of the answers. From any of the models.
Notion won 3-0.
The Result, Uncut
Here's the raw breakdown from foxish.ai/vs/alfred-vs-raycast:
| Model | Winner | Alfred mentions | Raycast mentions |
|---|---|---|---|
| ChatGPT | Notion | 0 | 0 |
| Gemini | Notion | 0 | 0 |
| Perplexity | Notion | 0 | 0 |
Two of the most beloved productivity tools in the Mac ecosystem, and the AI answer engines couldn't be bothered to name-drop either one when asked what a buyer should pick up right now. Instead, all three models pivoted to Notion, a product that isn't in the same category, doesn't solve the same problem, and won't help you launch an app with a keystroke.
Being loved by users isn't the same as being visible to models.That's the whole lesson in one line, but the reason it happened is worth unpacking. Because if you're a brand in this category and you're not Notion, this is your future unless you do something about it.
Why the Models Reached for Notion
The prompts we ran were shaped like real buyer intent. Things like "best productivity software to buy right now" and "top productivity tools for professionals." When a model encounters a prompt that broad, it does two things simultaneously: it infers what the user probably means by "productivity software," and it reaches for the highest-confidence brand associations it has in that semantic neighborhood.
Both of those steps hurt Alfred and Raycast.
Alfred and Raycast are launchers. The model sees "productivity." In the training data, the word "productivity" clusters overwhelmingly with note-taking, project management, and collaboration software. Notion. Asana. ClickUp. Todoist. Monday. The launcher category has its own semantic pocket ("Mac launcher," "Spotlight alternative," "keyboard-driven app launcher"), and unless the prompt uses one of those phrases, the model doesn't retrieve it. Notion has ambient PR gravity that the launchers don't. Notion has been the subject of hundreds of thousands of blog posts, LinkedIn threads, YouTube tutorials, Reddit debates, and productivity influencer roundups since 2019. Every "top productivity tools" listicle on the web mentions it. Every Y Combinator alumnus writes about their Notion setup. That corpus is the training data. Alfred, by contrast, gets referenced mostly inside niche Mac power-user communities. Raycast has more recent momentum, but most of its coverage lives in developer-focused publications and its own well-optimized docs. "Buy right now" language biases toward SaaS with pricing pages that look like SaaS. Notion has tiered plans, team seats, enterprise sales. Alfred sells a one-time Powerpack license. Raycast is largely free with a Pro tier. When the model interprets "buy," it tilts toward products where the purchase decision is legible and repeated across thousands of comparison articles.Add those signals together and the outcome isn't surprising. It's inevitable.
The Uncomfortable Part for Raycast
Raycast, in particular, should be uncomfortable with this result. The company has done a lot of things right on the surface: strong brand, gorgeous product, big developer community, a real content operation. If you Google "Raycast vs Alfred," you'll find a healthy mix of blog posts, Reddit threads, and YouTube reviews.
But Google visibility and AI visibility are not the same thing.
The Google SERP for "best productivity software" is full of listicles that mention Notion, Asana, and Trello at the top, with launchers appearing much further down (or not at all). When an LLM is trained on that corpus, it inherits that ranking as a prior. The model isn't looking at your beautiful landing page. It's looking at what other people said about your category, weighted by how often and how confidently they said it.
The AI doesn't visit your website. It reads what the internet says about you.Raycast's content is largely first-party: their own blog, their own docs, their own changelog. That's great for product marketing. It's not great for building the third-party citation graph that LLMs actually pull from.
What Alfred Could Do
Alfred's situation is different. It's not losing to Raycast in the model's eyes because both are invisible. But if we zoom out and ask "what would move the needle for Alfred specifically," the answer is less about beating Raycast and more about becoming legible to a model that currently has no idea it exists in a buying context.
Here's what actually moves numbers, based on the patterns we see across our scans:
Get into the listicles the model already trusts. When you inspect the sources ChatGPT and Perplexity pull from for productivity-adjacent queries, you see the same 20-30 domains over and over: Zapier's blog, The Verge, TechRadar, Tom's Guide, Wirecutter, PCMag, and a rotating cast of Medium and Substack writers with high domain authority in productivity. Alfred appears in some of these, but rarely at the top and rarely with a strong recommendation verb. Getting a pitch into a Zapier "best Mac productivity apps" roundup with a clear positioning line is worth more than a hundred product tweets. Publish comparison content on your own domain, but write it for the retriever, not the reader. A page titled "Alfred vs Spotlight vs Raycast: which macOS launcher wins in 2025" that gives an honest breakdown will get cited by Perplexity, which loves comparison content. It doesn't matter if you're biased — models learn to weight source credibility, but they also learn that first-party comparison pages exist and often quote them anyway when they're structured cleanly. Fix the category framing. Alfred's site describes itself as an "award-winning app for macOS which boosts your efficiency." That's brand copy. It's not retrieval copy. If the page opened with a sentence like "Alfred is a macOS productivity launcher and Spotlight replacement used by developers, designers, and knowledge workers to run apps, search files, and automate workflows with a single keyboard shortcut," you'd have a sentence a model can lift into an answer. Alfred's actual site copy is not that.What Raycast Could Do
Raycast is closer to breaking through than Alfred, but it needs to do two things it currently isn't doing at scale.
Show up in "productivity software" listicles, not just "Mac launcher" listicles. Right now, Raycast is well-indexed inside its own tight semantic pocket. If a user asks "best Mac launcher," Raycast probably appears. But the buyer prompts we tested were broader. Category expansion in the third-party corpus is a PR job, and it means pitching Raycast to writers covering productivity broadly, not just Mac tooling. Lean into the AI features publicly and loudly. Raycast AI is a legitimately good feature and it's the exact kind of product angle that generates coverage in the "AI productivity tools" content that's flooding the internet right now. Every one of those articles is training data for next year's models. Being in 50 "top AI productivity apps 2025" listicles would do more for Raycast's visibility in ChatGPT next year than any amount of paid acquisition this year. Build citation-worthy data. Raycast has usage data most companies would kill for. Publish a "State of Mac Productivity 2025" report with real numbers. Reports get cited. Citations become training data. Training data becomes recommendations.The Broader Lesson for Productivity Brands
If you're building anything in productivity software, whether it's a launcher, a habit tracker, a focus timer, a calendar app, a to-do list, a notes app, or a workflow automation tool, this matchup is a warning shot.
The category is dominated in the model's mind by three or four mega-brands: Notion, Asana, Trello, ClickUp, and sometimes Monday. If a buyer asks a broad question, those are the names that come out. Everyone else is fighting for scraps in the sub-queries — "best Mac launcher," "best focus app," "best Pomodoro timer" — where the semantic pocket is small enough that mid-tier brands can compete.
Two implications follow.
One: know which prompts you can actually win. You're probably not going to win "best productivity software." You might win "best keyboard-driven Mac launcher for developers." Map the prompt landscape for your category, figure out which ones you already show up in, and figure out which ones you have a plausible shot at. Fighting for prompts you can't win is a good way to spend money and see no movement. Two: the citation graph is your battlefield. Every third-party article, Reddit thread, YouTube video transcript, and podcast description that mentions your brand in a comparative or recommendation context is a training signal. Every one that doesn't is a gap. This is not the same as SEO, though it overlaps. It's not the same as PR, though it overlaps. It's a discipline of its own, and the brands that treat it as one are the ones that will be named when the next generation of models rolls out. The brands winning AI visibility in 2026 are the ones building citation graphs in 2025.What This Matchup Actually Told Us
The Alfred vs Raycast fight was supposed to be about two products. It ended up being about a category. Both brands got outflanked not by each other but by a product that isn't even a competitor, because the AI models don't distinguish as finely between sub-categories as humans do.
That's the pattern we see in matchup after matchup on our /vs tool. Two direct competitors get compared, and a third brand — usually the biggest brand in the parent category — walks off with the recommendation. It happens in project management (Monday and Asana lose to Notion). It happens in email (Gmail eats everything). It happens in CRM (Salesforce or HubSpot, depending on the prompt shape).
If you're a specialist tool competing in a category with a dominant generalist, the models will default to the generalist unless you give them a reason not to. That reason is a corpus of third-party content that positions your product in the specific slot you want to own.
Run Your Own Matchup
You can run any two brands against each other at foxish.ai/vs and see exactly how ChatGPT, Gemini, and Perplexity stack them. It's free. It takes about a minute. If you want the full picture on how your brand shows up across every prompt in your category (not just head-to-head), we do that too. Start a free scan →. No credit card, and the first scan is on us.