Comparison
Straight Edge GPT vs ChatGPT
ChatGPT is a generalist optimised to produce a good answer. Straight Edge GPT is a specialist optimised to produce a checkable one. Here is exactly where that difference shows up — hallucination.
Feature by feature
| Capability | Straight Edge GPT | ChatGPT |
|---|---|---|
| Citations on factual claims | Every claim carries an openable URL | Only when browsing triggers, often paraphrased |
| Confidence score | 0–100 with the basis stated in one line | None — tone is equally confident either way |
| Says "I don't know" | Required opening line below the confidence floor | Rare; usually produces a plausible answer instead |
| Source ranking | Filings, papers and raw data outrank blog summaries | Whatever the retriever surfaced, unranked |
| Contradiction detection | Conflicting sources shown side by side with the discrepancy | Conflicts silently averaged into one smooth claim |
| Tool-use transparency | Every query and page opened is shown in full | Collapsed or hidden |
| Exportable reasoning trace | Full JSON download, auditable by a third party | Not available |
| Speculative padding | "Some believe", "it could be" — banned unless requested | Common hedging filler |
| Ecosystem breadth | Focused on verified answering | Images, voice, apps, huge plugin ecosystem |
Comparison reflects default behaviour on typical research prompts. ChatGPT behaviour varies by model and whether browsing is triggered.
Four hallucinations, and what happens here instead
> How many people does the WHO estimate die from air pollution each year?
Typical generalist
"Around 7 million people die each year from air pollution." Stated flatly, no scope, no link.
Straight Edge GPT
7 million combined ambient and household; ambient alone is 4.2 million. Contradiction logged: two outlets quote different figures for the same claim because they use different scopes. Confidence 94/100, linked to the WHO fact sheet.
> What did the court rule in [an obscure 2023 case]?
Typical generalist
Invents a plausible holding, a plausible judge and a plausible docket number. All three can be fabricated.
Straight Edge GPT
"I don't know the answer to that, but here's what I think…" — searches, finds no primary record, declares the gap, and refuses to name a holding it cannot open.
> Cite three studies showing X works.
Typical generalist
Returns three citations, one or two of which do not exist or don't say what's claimed.
Straight Edge GPT
Opens each paper before citing it. Anything it could not retrieve is dropped, and the answer says how many candidates failed verification.
> What's the current price / latest release / who won last night?
Typical generalist
Answers from a training cut-off unless browsing kicks in, with no warning that it's stale.
Straight Edge GPT
Anything checkable is checked live at answer time, and the retrieval timestamp is part of the verdict.
Test it on a question you already know the answer to
That's the honest benchmark. Three free answers, no card, no account. Then $19.99/mo, or $9.99/mo billed yearly.