Factually: AI-powered search for reliable answers

Last week, Factually appeared on Hacker News with barely a point and no comments, which is not necessarily an indicator of quality, but does prompt the question of whether there is something here worth paying attention to. At a time when nearly any wrapper around an LLM is marketed as "the search engine of the future," it is worth examining more carefully what this project actually proposes.

Factually positions itself as an AI-powered research tool designed to find reliable answers. The core proposition is to reduce the verification burden placed on the user when querying a language model. Instead of receiving plausible text without clear sources, the system aims to anchor its responses in concrete, verifiable references.

What it actually does

According to information on its website, Factually combines real-time web search capabilities with structured answer generation, prioritizing information traceability. Each relevant claim is linked to its original source, so users can audit the system's reasoning without relying solely on trust in the underlying model.

The workflow is straightforward: enter a question or research topic, and the tool returns a summary with explicit citations. This is not a new concept—Perplexity AI has been working this same territory for several years, and tools like You.com or Bing Copilot include similar functionality—but the concrete execution and editorial approach of each product create differences that only become apparent with sustained use.

Why this approach matters

The problem of hallucinations in LLMs has not disappeared with the latest generations of models. Even with large context windows and reasoning improvements, systems remain capable of generating incorrect claims with apparent confidence. For research tasks—journalism, legal analysis, due diligence, academic studies—that uncertainty carries real cost.

Tools that prioritize verifiability over narrative fluency respond to a legitimate need. Not every user wants an elaborated answer; sometimes you need to know where each piece of data comes from. In that sense, any product that prioritizes epistemic transparency over the appearance of omniscience is addressing something structurally useful.

Who it makes sense for

The most obvious user profile is researchers, analysts, or journalists who need verifiable backing for their work. It is also relevant for teams tracking complex subjects and requiring information traceability over time.

For the casual user who simply wants a quick answer, the proposition may feel slower or denser than necessary. Factually's differentiating value, if it ultimately has one, lies precisely in that additional rigor, which for some workflows is essential and for others, superfluous.

What we still don't know

The limited initial traction on Hacker News—one point, zero comments at publication—does not allow conclusions about the product's actual quality. It could be a quiet launch of something solid, or it could be a very early-stage project. Without access to extensive testing or technical documentation about the model or search indices it uses, any deep assessment would be premature.

What stands out is the absence of clear information about the technical stack. There is no specification of which underlying model powers the responses or how the source index is updated. For a tool whose value proposition is precisely reliability, this opacity is something that needs addressing.

---

From ClaudeWave, we will follow Factually's evolution with measured interest. The niche of AI-powered verified search has real demand, but also several already-established competitors. Time will tell whether the execution lives up to the promise.