Fan-out and prompt tracking,
benchmarked.

Two features now separate a serious AI-search tracker from a dashboard with a logo: how it surfaces query fan-out (the hidden sub-queries an assistant runs before it answers) and how it does prompt tracking (which prompts it watches, how often, and across which models). We read the first-party documentation for five tools and lined them up against each other. The headline: the tools agree fan-out matters, but they disagree on what to call it, whether it's real or synthetic, and how openly they document it.

What query fan-out actually is

When you ask an assistant something that needs live information, it rarely runs a single search. It decomposes your prompt into a handful of sub-queries — different angles, attributes, comparisons — runs each as its own search, and synthesizes the answer from all of them. Those sub-queries are the fan-out. If your brand ranks for the prompt but not for the specific sub-queries the model generated, you can be invisible in the final answer anyway. That's the gap every tool in this benchmark is trying to expose. (For the mechanics and how to read fan-out queries yourself, see our guide on finding ChatGPT's hidden fan-out queries.)

How Profound surfaces fan-out

Profound ships fan-out as a first-class, named feature — "Query Fanouts" — visible per tracked prompt. Below, a Profound account breaks a single prompt into its fan-out variations, with a query count and an average number of queries per execution. One prompt ("hardware and furniture wholesalers") expands into 83 fan-out variations and 163 captured queries.

The same surface aggregates across a brand's whole prompt set. Here an aerospace-supplier account shows its top prompts ranked by query count — "top aerospace CNC machining companies" alone generated 149 captured queries across 128 fan-out variations, at 2.9 queries per execution.

Profound describes these as the "actual, high-intent queries" captured from prompts it sends every day — so the fan-out is real / longitudinal, not modelled. It pairs that with the strongest documented real-conversation prompt sourcing in this group: Prompt Volumes validated against a 1.3B+ conversation dataset and Prompt Research Reports built on 1.5B+ real user prompts.

How Peec AI surfaces fan-out

Peec is the only vendor here documenting two fan-out surfaces. The first is real and lives inside each chat. In Peec's "Anatomy of a chat," Fanout Queries sit alongside the main response, brands mentioned, and the sources sidebar — "the query fanouts performed by the model when generating the answer for a specific prompt."

The second view is per-prompt. On an individual prompt dashboard, Peec shows the latest tracked Query Fanouts plus the most common terms across them — "what similar queries the model performed while creating an answer for the specific prompt." Note the caveat in the docs: this surface is labelled "for ChatGPT only."

On top of the captured surface, Peec launched a separate Synthetic Query Fanout in Early Access (8 May 2026), which predicts fan-outs rather than capturing them — ChatGPT only at launch. That makes Peec the cleanest example in the group of a vendor splitting real and synthetic fan-out into two products.

Fan-out, side by side

Here's the full fan-out comparison across all five tools, drawn from first-party documentation as of June 2026.

A recurring caveat worth flagging: four of the five tools do not clearly document an inline Google/Bing rank view next to fan-outs. Semrush is the exception. Its announcement explicitly says Query Fan-Out Analysis reveals the Google search queries, the domains ranking for each query, and where brands are missing from high-impact SERPs. Ahrefs is the only vendor here that publishes a typical returned count: most often two fan-outs per prompt, with one or none for simple prompts.

Prompt tracking, side by side

Fan-out is only half the story. The other half is which prompts you track and how. The tools split into three patterns: huge first-party conversation datasets for prompt discovery (Profound, and to a degree Peec and Semrush), workflow-centric generation tied to personas and funnel stages (Scrunch), and a giant search-backed prompt index alongside custom prompts (Ahrefs).

One gap is shared by every vendor: none of their first-party pages clearly documents a standard "run each prompt N times and average the outputs" method for handling AI-response volatility. Several acknowledge the variance and steer you toward trends over multi-week windows instead — which is exactly the problem we dug into when we ran 19 prompts through ChatGPT ten times each and found only 16% of recommended brands were stable across all ten runs.

Where the documentation conflicts

If you're evaluating these tools, the conflicts matter as much as the features — they tell you what's genuinely shipped versus marketed:

• Peec fan-out coverage. The docs say prompt-level Query Fanouts are "for ChatGPT only," but Peec's May 2026 blog says the dashboard shows what "ChatGPT, Grok, and Perplexity searched."
• Scrunch plan names. Pricing lists Starter/Growth/Enterprise; the free-trial FAQ lists Explorer/Core/Enterprise/Agency Core. Platform count is mostly "eight, Grok soon," with one page implying nine.
• Ahrefs custom-prompt coverage. The current help page lists Grok alongside ChatGPT, Perplexity, Gemini, and Copilot, but says Ahrefs cannot collect new Grok data for now; its plan-allowance examples still enumerate only the four non-Grok assistants.
• Semrush engines. Three first-party pages give three different Prompt Tracking engine lists. Country coverage is documented as both 32 countries (May 2026 news) and 15 regions (older toolkit page).

How Aiso fits

Aiso comes at the same problem from the demand side. Instead of only replaying your own test prompts, we surface the real prompts customers ask AI assistants from a consent-based panel, the fan-out queries those assistants run behind the scenes, and the citations they return — so you can see where you're missing across the sub-queries that actually decide the answer, not just the headline prompt. If you want the manual version first, our walkthrough on finding fan-out queries the hard way shows what these tools are automating.