On 29 July 2021, the U.S. Attorney’s Office for the Southern District of New York charged Trevor Milton, Nikola Corporation’s founder and former chief executive officer and executive chairman, with two counts of securities fraud and one count of wire fraud concerning statements made between November 2019 and September 2020 about Nikola’s products and technology.Reuters reported that Milton pleaded not guilty and was released on a US$100 million bond, while Nikola was not charged in the criminal case and stated that it was cooperating with the government.The criminal case concerned alleged efforts to induce retail investors to purchase Nikola shares through public statements disseminated through social media and television, print and podcast interviews; according to the U.S. Attorney’s Office, the statements concerned Nikola’s product and technology development.
How does Writfind compare with ChatGPT Deep Research and Gemini Deep Research?
3 companies, same investigation scope, 76 reference matters. Every report was graded blind against the same set.
Three tools, one reference set
Recorded pooled recall: Writfind 73.7%–85.5%; Gemini Deep Research 48.7%; ChatGPT Deep Research 42.1%–43.4%. The records cover 3 graded Writfind report versions; each Deep Research report was generated once. These ranges reflect report revisions and rescoring, not independent repeated trials or confidence intervals. The table and charts show the latest published comparison.
Writfind conducted this comparison in September 2026 on 3 development cases. The 76-matter reference set includes five matters from earlier research. Unknown matters missed by every investigation cannot be counted; these cases do not establish performance on other companies.
Tool
Recall
Precision
F1
Evidence support
Writfind
73.7%56 of 76
96.6%2 recorded rejections
83.6%
99.4%160 of 161
Gemini Deep Research
48.7%37 of 76
100%0 recorded rejections
65.5%
93.3%140 of 150
ChatGPT Deep Research
42.1%32 of 76
100%0 recorded rejections
59.3%
91.2%145 of 159
Magenta marks the best score in each column.
Precision and F1 are calculated from the recorded comparison counts. Writfind’s two rejected candidates were carried over from an earlier report version; its current-report false-positive count has not been re-assessed.
Further right means more reference matters found. Higher means more evaluated passages have source support for their central claim.
Writfind73.7% / 99.4%
Gemini Deep Research48.7% / 93.3%
ChatGPT Deep Research42.1% / 91.2%
Recall / evidence support
100%Evidence support0%
Writfind
Gemini
ChatGPT
0%25%50%75%100%
Recall
When a source fails, does the report say so?
The evaluation recorded passages without confirmed source support, missing citations and inaccessible source pages. The sample reports also mark passages based only on search snippets.
Tool
Support not confirmed
Passages without citations
Unreachable source URLs
Sources marked unread
Writfind
1
0
0
6
Gemini Deep Research
10
25
2
Not recorded
ChatGPT Deep Research
14
42
3
Not recorded
Totals across the three evaluated reports for each tool. The first two columns count passages; the last two count sources.
What the citations point to
Writfind · 218 cited source pages
Gemini Deep Research · 91 cited source pages
ChatGPT Deep Research · 51 cited source pages
Official record
Major media
General media
Portal
Commercial database
Social media
The three reports behind the numbers
Nikola Corporation
Flagged
Battery-electric and hydrogen trucks. Delaware; formerly Nasdaq: NKLA.
In our comparison of Nikola, Ozy Media and Binance, general AI deep research tools provided useful leads. Writfind covered more reference matters overall; Gemini led on Nikola. The results support initial discovery, but decisions about important matters still need checks on developments, omissions and sources.
A due diligence manager reviewing a supplier needs to know whether the record shows repeated misconduct, an unresolved allegation or an issue addressed by later evidence. Generative AI can help organize the material, but a visual summary can make those alternatives appear settled before the sources justify them. Each graphic should show what judgment the evidence supports and what information could change it.