The citation is the new click, and most creators are still optimizing for the wrong one
If you publish for a living, the ground under your distribution has quietly shifted. The buyer, the follower, the person who used to land on your site from a Google search — they’re increasingly asking ChatGPT, Gemini, Perplexity, or Claude first, and those models answer without ever sending a session your way. That means your visibility is now decided inside a black box you can’t inspect with Google Analytics. A new tool called Prefer, built by the team at Prefer, is trying to sell creators and brands a way to act on that gap rather than just stare at a dashboard about it. Whether you’re a solo creator or running social for a brand, the underlying shift — from ranking to being cited — is the one to understand this quarter.
The problem is not measurement, it’s the missing second half of the loop
The maker, Javed Khatri, frames it bluntly in the launch thread: after 13 years working in SEO, his team noticed that almost every AEO tool is good at showing numbers — your AI visibility score, your share of voice, how often you’re mentioned, how you stack up against competitors — but “users do not really want numbers.” That’s a sharp observation, and in my experience running content for small brands it’s correct. Knowing you’re invisible in AI answers is not the hard part. Knowing which page to rewrite, which third-party source to pitch, and whether any of it moved the needle is the hard part.
His thesis is that “measuring the gap is the easy half. Closing it is the product.” Prefer positions itself as the closing half: it claims to do the on-page, content, and off-page work needed to win a citation, with the pitch being “less dashboards, more action.” That’s a deliberate contrast with the wave of AEO trackers that have flooded the market — think the reporting-layer tools that sit next to Ahrefs or Semrush and spit out an “AI visibility” number. Whether that contrast holds up in practice is the question I’d want answered before I paid for either category.
Why this matters more to creators than to enterprise SEO teams
Here’s the part the launch copy undersells. If you’re a social media manager or indie founder, you already do the “action” half by instinct — you write, you publish, you pitch, you iterate. What you don’t have is a feedback loop telling you whether the model cited your blog post, your YouTube description, or that Reddit thread you seeded. So a tool that closes the loop is theoretically more valuable to a lean operator than to a big brand with an agency. The catch is that lean operators are also the least likely to pay enterprise prices, which is where the Claude detail below gets interesting.
What Prefer actually claims to do, and where the claims get fuzzy
The core mechanic, per the maker’s replies, is a continuous loop: establish a baseline for “the prompts and queries that matter,” re-run them across ChatGPT, Claude, Gemini, and Perplexity, take an action, then track what changes afterward. Specifically, the team says it watches whether the brand gets mentioned or recommended, which sources start getting cited, which competitors gain or lose visibility, whether the model changes its preferred sources, and whether the improvement persists across multiple runs. If there’s no movement, the system loops back and identifies what’s still missing.
That’s a reasonable-sounding design, and it directly answers the sharpest question in the thread, from Sahil Kathpal, who asked how you separate real lift from noise given that model answers “change a lot between runs.” The maker’s answer — don’t treat a single model response as the metric, track changes over time — is the right instinct. It’s also exactly the kind of claim that’s easy to assert and hard to verify from the outside.
The honesty problem with “AI visibility” scores
I want to flag something as opinion, not fact: any AEO score is a proxy, and proxies drift. When I’ve tested similar trackers, the same prompt run twice an hour apart produced different source lists, and the “share of voice” number swung by double digits without anything changing on the site. That’s not a knock on Prefer specifically — it’s a structural problem with the whole category. The maker’s insistence on persistence-over-multiple-runs is the correct mitigation, but it also means you can’t judge the tool from a single screenshot. You’d need weeks of data. Treat any vendor’s before/after chart, including this one’s, as marketing until you’ve run your own baseline.
What social media operators can borrow from this, tool or no tool
You don’t need to buy anything to apply the logic, and honestly, most creators shouldn’t yet. The transferable playbook is simple:
- Pick 10–20 prompts your audience actually types. Not vanity queries — the real ones. “Best [your niche] newsletter,” “how to [thing you teach],” “[your category] alternatives.” Run them manually across ChatGPT, Perplexity, Gemini, and Claude once a week. Screenshot the answers.
- Log which sources get cited. In my experience, models lean hard on a small set of domains: Reddit, Wikipedia, YouTube transcripts, a handful of authority blogs, and increasingly LinkedIn posts. If those are the citation wells, that’s where your off-page effort goes — not into a new landing page.
- Treat your owned content as citation bait. Structured, quotable, dated, with clear claims and a named author. That’s the same discipline that made you rank in 2015, repackaged.
- Repurpose across surfaces the models read. A YouTube video with a full transcript, a LinkedIn post that restates the thesis, a Reddit comment that answers the question directly — each is a separate citation opportunity. Tools like Repurpose.io or Opus Clip help you multiply the raw material; the models reward the redundancy.
Why TikTok creators should care more than LinkedIn ones
Counterintuitive, but I’d bet on it. LinkedIn gets cited constantly by AI models because it’s text, indexed, and full of named professionals. TikTok is video-first and far less machine-readable, so its creators are effectively invisible to AI answers today — which means the TikTok creator who also publishes a written version of their best hooks (a blog post, a newsletter, a Threads thread) is arbitraging a gap. The video builds the audience; the text builds the citation. If you’re only doing video, you’re building a moat the models can’t see over.
Where my judgment says Prefer falls short
Two things stand out, and both come straight from the thread rather than from me.
First, the Claude gap. Gene Dai pointed out that the launch copy leads with Claude, but the pricing page lists Claude Sonnet 4 only via API on Enterprise — not on self-serve tiers. The maker, Mandar Sawant, confirmed they’re “working towards adding Claude models to the self serve tiers” and kept it enterprise-only “over api for exactly the reason you mentioned” — namely that API answers can differ from what people see in the app. That’s a candid answer, and it’s also a real limitation: if you’re a solo creator on a self-serve plan, you may not get the Claude coverage the marketing implies. Pricing itself is not disclosed in the source, so I can’t tell you whether it’s a bargain or a stretch.
Second, the support signal. Jordan Gibbs reported being in “softlock” — unable to load a page without a failure — and said the support email “doesn’t work either.” That’s one data point from an early-stage, bootstrapped product, and early products break. But for an operator deciding whether to route your visibility workflow through a new tool, a broken support channel is the kind of thing that should make you keep your old process running in parallel for a quarter. I’m not calling it a pattern; I’m saying one report plus “bootstrapped and early” equals “don’t make this your single point of failure.”
Who this is not for
If you’re a creator whose entire business is a single platform’s algorithm — say, pure TikTok reach monetized through the Creator Fund — Prefer solves a problem you don’t have yet. AI citation matters when your funnel includes a website, a newsletter, or a product page that a model could recommend. If your money comes from the For You page, spend your energy on hooks and watch time, not AEO. Similarly, if you’re a large brand already running an agency-led SEO program, this is a “test alongside” tool, not a replacement — the category is too young for anyone to rip out their existing stack.
The bigger trend this launch is riding
Step back and the pattern is obvious: every distribution channel eventually gets an optimization layer, and the layer arrives before anyone agrees on how to measure it. We watched this with social scheduling — Buffer, Hootsuite, Later, and Metricool all sold “post everywhere from one place” long before anyone could prove cross-posting drove revenue. We watched it with analytics, where UTM tracking became gospel despite last-click being a known lie. AEO is the same movie with a new cast: ChatGPT, Gemini, Perplexity, and Claude are the new distribution surfaces, and the tooling to game them is being built in public, right now, on Product Hunt.
My take: the winners in this category won’t be the ones with the prettiest score. They’ll be the ones whose actions demonstrably move citations, and who are honest about how noisy the signal is. Prefer has staked out the right position — action over dashboards — and the maker’s replies show they understand the noise problem. Whether the execution matches the pitch is something only a few weeks of your own baseline data can tell you.
What I’d watch / test next
This week, before you spend a dollar on any AEO tool, do the free version of the loop. Open ChatGPT, Claude, Gemini, and Perplexity, and run the same 10 prompts a prospect would use to find someone like you. Paste the answers into a doc, highlight every source cited, and note whether your name or your brand appears. That’s your baseline — and it costs you an hour. Then publish one piece of structured, quotable, text-based content that directly answers the most common prompt, and re-run the same 10 prompts in two weeks. If nothing moves, you’ve learned the category is harder than the vendors admit. If something moves, you’ve learned which surface the models actually read — and that is the insight worth paying for, from Prefer or anyone else. Watch the Claude self-serve rollout and the support reliability; both are the tells for whether this is a tool you build a workflow around or one you keep at arm’s length.
