Our stance.
Published September 24, 2026 — this page states what itclicks refuses to do, and what it promises instead. It changes only if either one does.
Ask a general-purpose AI assistant a question with more than one part to it, and watch what happens: it takes several separate facts, pulled from different places, and folds them into one confident-sounding sentence. If there are citations at all, they get attached loosely — a footnote at the end, a source list underneath — but nothing tells you which specific claim came from which specific source. Four facts. One sentence. One blended answer.
We call that “Smoothing.”
Smoothing isn't a bug. It's what happens when a system is built to sound complete rather than to stay checkable. It reads well. It sounds sure of itself. And there's no way, from inside that sentence, to tell where any one piece of it actually came from.
itclicks refuses to do that.
If a question has four separate answers folded into it, itclicks gives you four separate points — one at a time, each one holding its own source, each one you can check before you move to the next. That's slower and more deliberate than getting one smooth paragraph back. It's an intentional tradeoff: the moment itclicks starts merging separate facts into one confident sentence just to feel faster, it's no longer checkable — and it's no longer itclicks.
We don't smooth it over.We keep it checkable.
That's the whole stance. The pacing, the sourcing, the refusal to guess when nothing checks out — all of it comes from that one line.
We didn't coin the problem
We named it “smoothing” for plain English, but the pattern itself is documented and measured. Three places that back this up, in their words, not ours:
Audited four AI search engines by hand for whether citations actually supported what they were attached to.
“…on average, a mere 51.5% of generated sentences are fully supported by citations and only 74.5% of citations support their associated sentence.”View aclanthology.org ↗
Names the failure: citations that line up with what the model already believed, not with what it actually relied on.
“Faithfulness ensures that the model's reliance on cited documents is genuine, reflecting actual reference use rather than superficial alignment with prior beliefs, which we call post-rationalization.”View arxiv.org ↗
Compared eight AI search engines on how well they cite the news.
“Collectively, they provided incorrect answers to more than 60 percent of queries.”View cjr.org ↗
“Every sourced answer is backed by a word-for-word quote from that exact source, checked by machine before you ever see it — not asserted, verified. The rare answer that isn't sourced (a plain dictionary-level definition) is labeled exactly that: general knowledge, not checked against a source.”