Key takeaways
- Query fan-out is the process by which ChatGPT, Google AI Mode, and Perplexity silently expand one prompt into several parallel sub-queries before composing an answer. Most keyword tools never see this layer.
- Only a handful of platforms show the real sub-query tree pulled from live monitoring. Everyone else sells a simulator that guesses what a model might do.
- Profound, Otterly.AI, LLMrefs, and Semrush's Enterprise AIO tier are the four names that come up repeatedly when people ask for actual fan-out visibility, each with a very different depth and price.
- Free simulators (Qforia, Rankability, Wellow) are fine for a one-off content brief but useless for ongoing tracking since fan-out patterns shift week to week.
- Promptwatch treats fan-out as one input among many, pairing it with crawler logs and citation data so you see not just the sub-queries but whether your pages are actually winning them.
Why the sub-query tree matters more than your rank tracker
Here's the thing nobody tells you when you start chasing AI visibility: the prompt your customer typed is not the query that gets searched. Google confirmed this directly when describing AI Mode, saying the system "breaks down your question into subtopics and issues a multitude of queries simultaneously on your behalf." ChatGPT does something similar. Ask it "best CRM for a 20-person B2B SaaS startup" and behind the scenes it's quietly running separate retrieval passes for pricing, HubSpot integrations, G2 reviews, and API availability, then stitching the best passages from each into one answer.
Miss one branch of that tree and you never existed for that answer, no matter how well you rank for the original phrase. According to Promptwatch's research on ChatGPT query fanouts, the average response now triggers somewhere between 1 and 2 separate web searches, though that number has swung wildly over the past year. In early December it averaged 2.15 queries per response; by March it had dropped to 1.84; after a data gap, April came back at a much leaner 1.0 per response. The queries themselves got shorter too, falling from roughly 117 characters in December to about 53 characters by April, meaning ChatGPT increasingly searches like someone typing keywords rather than asking a full question.
Then in August, something changed again. Promptwatch's data on ChatGPT's site: operator shows that on August 8, 2026, ChatGPT Search overnight jumped its use of the site: operator in fanout queries from roughly 0.37% to 16.8%, a 46x increase in a single day, while average fanouts per response nearly doubled from 1.08 to 1.83. That means ChatGPT is now directly running site:yourdomain.com searches against your own website as part of answering someone else's question. If your internal search is broken or your category pages are thin, you're losing citations you never knew you were competing for.
This is the layer that separates a genuine AI visibility platform from a glorified keyword tool wearing an AI costume. So which ones actually show you the tree, versus just implying one exists?
The platforms that show a real fan-out tree
Profound: the deepest feature set, gated behind enterprise pricing
Profound's Query Fan Out feature visualizes every sub-query variation generated from a tracked prompt, complete with a share percentage showing how often each variation gets used to fetch results. It layers on three more things worth knowing about: word transformations (which words the model adds, drops, or keeps when rewriting your prompt), freshness detection (flagging date tokens that signal a preference for recent content), and persona impact (how the fan-out tree shifts depending on the buyer persona or region you're simulating).
Profound frames all of this around what it calls the "three gates of AI retrieval": is your content fetchable, is it chosen, and is it extractable. It's a genuinely useful mental model.
The catch is pricing. Profound pulled its self-serve Starter and Growth plans in mid-September 2026. What's left for individual brands is a free trial (10 prompts, ChatGPT only, run once) and a custom-quote Enterprise tier that covers up to nine engines. Agencies can get in at $99/month plus $399/month per client workspace. If you want the real fan-out feature set and not just a taste, you're likely negotiating an enterprise contract.
Profound
Otterly.AI: free fan-out comparison across three engines at once
Otterly.AI's free Query Fan-Out tool runs one prompt across Google AI Overview, Google AI Mode, and ChatGPT simultaneously and shows where the expansions diverge. It's a smart pitch, framed around the idea that "your page isn't competing for the query someone typed, it's competing for eight or twelve sub-queries you never see." Otterly sorts the sub-queries into buckets: comparisons, qualifiers like "for beginners" or "under $50", follow-ups, and adjacent intents.
The free tool is genuinely handy for a quick gut check. Paid plans start at $29/month for ongoing tracking, with enterprise pricing starting around $1,000/month. Third-party comparisons classify Otterly's fan-out depth as shallow to medium, without white-label reporting or a full API, which fits its positioning as a lighter tool for smaller teams.
Otterly.AI
LLMrefs: honest about the difference between simulated and real fan-out
LLMrefs does something the others don't: it explicitly tells you the difference between its free Query Fan-Out Generator, which simulates sub-queries using a model fine-tuned on historical prompt data, and the real fan-out data that only shows up in its paid monitoring, where you see the actual sub-queries real users triggered and whether your content showed up against competitors.
LLMrefs also breaks down how fan-out behaves differently by platform: AI Overviews runs a lighter version for quick answers, AI Mode runs an aggressive version for complex queries (and literally displays "Searching 8 queries" in its UI), and Perplexity shows its fan-out most transparently of the three, tying inline citations directly to specific sub-queries. The paid plan, currently $79/month at a limited-time rate, tracks 500 prompts for mentions, sources, and fan-out queries.
LLMrefs
Semrush: real Google fan-out data, but only on Enterprise
Semrush splits its fan-out capability into two tiers. The AI Visibility Toolkit's Prompt Research report approximates likely sub-queries using a "Related Topics" feature, which is closer to a simulator than real monitoring. The actual sub-query data, Google's genuine fan-out queries at scale with ranking positions and ranking domains, lives inside Enterprise AIO's Query Fan-Out Analysis automation, and that's Enterprise-only.
Semrush ran its own optimization experiment updating four blog articles to target fan-out queries. Citations rose from 2 to 5 over a month, a 150% increase that peaked at 9 before settling. But Share of Voice actually fell from 23.4% to 20.0% over the same window, and brand mentions dropped from 18 to 10, a decline Semrush attributes partly to AI platforms mentioning fewer brand names overall during the test period, since competitors like Ahrefs and HubSpot saw similar drops. It's a useful reminder that a citation increase doesn't automatically mean a visibility increase, and that fan-out optimization results are genuinely hard to isolate from platform-wide noise.
The free simulators: good for a one-off brief, useless for tracking
A cluster of free tools exist purely to simulate what a fan-out tree might look like for a given prompt. They're worth knowing about because they cost nothing, but none of them monitor anything over time.
Qforia, built by Mike King's team at iPullRank, is Gemini-powered and simulates fan-out for both AI Overviews and AI Mode, though you need your own paid Gemini API key to run it. Rankability's free AI Search Query Fan-Out tool is solid for mapping query branches and generating a content brief in one pass, but it doesn't track anything afterward. Wellow's generator does semantic query expansion and intent grouping from a single keyword, again with no ongoing monitoring layer.
The problem with all three is that fan-out patterns aren't static. Citation rates on the same tracked prompts can swing 40-60% month to month as models get updated, so a snapshot from March tells you almost nothing about September. If you're serious about this, you need something that re-checks the tree on a schedule, not a tool you run once before writing a blog post.
Comparison table: fan-out reporting depth by platform
| Tool | Fan-out type | Engines covered | Ongoing monitoring | Price entry point |
|---|---|---|---|---|
| Profound | Real, with share %, word transforms, freshness, persona | Up to 9 (Enterprise) | Yes | Custom quote (agency entry $99/mo + $399/client) |
| Otterly.AI | Free comparison tool + paid real tracking | ChatGPT, AI Overviews, AI Mode (free); 7 engines (paid) | Yes, paid tier | Free tool; paid from $29/mo |
| LLMrefs | Free simulator + real tracked sub-queries | ChatGPT, Perplexity, 9 total engines | Yes, paid tier | $79/mo |
| Semrush Enterprise AIO | Real Google fan-out at scale | Primarily Google surfaces | Yes | Enterprise, custom-quote |
| Qforia | Simulated only | AI Overviews, AI Mode | No | Free (own Gemini API key required) |
| Rankability | Simulated only | Google AI surfaces | No | Free |
| Wellow | Simulated only | Generic | No | Free |
Why fan-out visibility alone isn't the whole picture
Here's my honest take after going through all of this: knowing the sub-query tree is necessary but not sufficient. Seeing that ChatGPT splits your prompt into eight sub-queries tells you what the model is looking for. It doesn't tell you why your content isn't showing up in the answer, or whether AI crawlers even reached your page before the model needed it.
That's the gap Promptwatch is built to close. It tracks prompt volumes, difficulty, and query fan-outs alongside crawler logs that show exactly when ChatGPTBot, ClaudeBot, PerplexityBot, and 400+ other bots hit your site, what they read, and where they errored out. Pair that with citation trend data classified across 22 content types, and you can connect the dots between "this sub-query exists" and "this specific page either got cited or got skipped, and here's why." Given that Promptwatch's own data on average sources per response shows ChatGPT only cites around 5 sources per answer, roughly half of Google AI Overviews' 10, every fan-out branch you lose really does cost you a scarce citation slot.
Most of the fan-out-specific tools above stop at the tree itself. They'll show you the branches; they won't tell you whether your CMS is publishing content fast enough to compete for the newer, shorter, keyword-style queries ChatGPT has been running since its character lengths dropped to roughly 53 characters this spring. That's a content operations problem, not a monitoring problem, and it's where the fan-out trackers hand off to platforms with actual content agents attached.
How to pick, practically
If you're a solo operator or small content team testing the waters, start with Otterly's free comparison tool or LLMrefs' simulator. Neither costs anything and both will show you, in about five minutes, that your target prompt splits into more sub-queries than you assumed. That alone changes how you write the next article.
If you're running content for an agency managing multiple client brands, LLMrefs at $79/month or Otterly's paid tier at $29/month are the more affordable entry points with real ongoing tracking, though neither matches Profound's depth of feature.
If you're at a company where AI-driven citations already move revenue, and you need the freshness detection, persona impact, and word-transformation layers on top of the raw tree, Profound's enterprise tier is the most complete fan-out feature set available, assuming you can absorb the pricing.
If what you actually want is to close the loop between "here's the sub-query tree" and "here's the content we published to win it, and here's proof AI crawlers read it," that's a different category of tool entirely, and it's worth browsing the wider field of options in the GEO software directory at bestgeosoftware.com before committing to a single-purpose fan-out tracker.
The honest bottom line
Query fan-out reporting got noisy fast in 2026 because every AI visibility vendor realized "we show sub-queries" is a good sales line. Fewer of them actually monitor real fan-out trees over time; most sell you a one-time simulation dressed up as ongoing intelligence. Profound, Otterly, LLMrefs, and Semrush's Enterprise AIO tier are the four that genuinely pull live sub-query data rather than guessing at it. Everything else on the free-tool list is worth five minutes of your time and nothing more.