100 Password Manager Questions: What ChatGPT Recommends

The password manager ChatGPT recommended most often was Bitwarden, in 82 of 100 answers, and Bitwarden was its own pick in 78 of the 100. In short: one default, a vendor-written reading list, and a fallen leader.

The strongest default in the series

Bitwarden is the default, and the strongest the series has seen: named in 84 of 100 answers, recommended in 82 and picked in 78. No brand in the nine earlier teardowns reached any of those numbers; the previous highs were 75 named (QuickBooks and HubSpot), 74 recommended (HubSpot) and 70 picked (Asana). Eighty of the 100 questions do not mention Bitwarden, and it is picked in 60 of those answers. 1Password is a clear second, named in 67, recommended in 65 and picked in 59. The two are picked together in 49 answers, typically 1Password for the most polished experience and Bitwarden for price, open source and self-hosting. Of the 93 answers that pick anything, 88 pick one or both.

Everyone else is a lane

Below Bitwarden and 1Password, every brand has its lane: Proton Pass is named in 38 answers, recommended in 36 and picked in 27, usually for privacy. Keeper is picked in 21 of the 25 answers that name it, 14 of them to questions framed around a business or team, and in all four healthcare, law firm and MSP questions that produced a pick. KeePassXC is picked in 16 of 22, for offline, local-file control, 11 of them in the 20 alternatives questions. Dashlane has the widest gap between being described and being chosen: recommended in 26 answers, picked in 15. Apple Passwords is recommended in 14 and picked in 8, almost always for people who use only Apple devices.

LastPass has almost disappeared

Unless the question asks about it, LastPass almost never comes up. Eleven of the 100 questions name LastPass. In the other 89 answers it is named twice: once in a healthcare table and once as a cited blog. When a question does ask about it, 6 of the 11 answers bring up its 2022 security incident. In the six head-to-heads that name it, LastPass is picked in three, each time on a condition such as wanting a free option or already being invested in it; the other side is picked in five, and one answer picks neither. For one of the best-known names in the category, that is a striking absence, and a sign of how a reputation event can carry into AI answers.

Decisive answers, from sites that do not rank

ChatGPT commits to a pick in almost every answer, usually for a brand whose own site does not rank for the question. Ninety-three of the 100 answers pick at least one brand, and 27 of the 30 head-to-heads pick every brand named. Nineteen of the 20 direct advice questions produced a pick, with Bitwarden among the picks in 18 and 1Password in 17; the one without a pick, for a healthcare practice, asked how many users the practice has and which office suite it runs, and named no product. Of the 241 picks, 148 went to brands whose own website did not rank in Google's top ten for that question, 61 percent. Bitwarden's site ranked for 43 of its 78 picks, 1Password's for 19 of its 59. Google's top ten is 84 percent third-party pages, level with applicant tracking as the highest of the first ten categories.

The vendors write the reading list

Vendors supply most of the reading list. Across 100 answers there were 292 citation events to 77 domains. Vendors' own sites took 201 of them, 69 percent, the highest share of the first ten categories, and 65 answers cited nothing but vendor pages, the most of any category in the series. bitwarden.com alone is cited in 64 answers, 45 of them to questions that never mention Bitwarden, and Bitwarden is picked in 41 of those 45. 1password.com is cited in 52 answers and proton.me in 28. The most-cited independent sources are Wired (10) and security.org (8). Reddit and Wikipedia have zero citations, for the tenth teardown running.

One hundred three of the 292 citation events involved a domain that also sat in Google's top ten for that query, 35 percent.

The table sets ChatGPT's password manager answers beside the other categories tested earlier in the series, side by side under the same rulebook.

What ChatGPT's recommendations mean for password manager brands comes down to four things, in order.

Your own site can be the category's textbook. bitwarden.com was cited in 45 answers to questions that do not mention Bitwarden, and Bitwarden was picked in 41 of them. Clear comparison, pricing and security pages are the material the model reaches for first in this category.

Reputation events show up in the answer. LastPass is still a working product with business customers, but the model mentions it almost only when asked, and then with its breach. Recovering from an event like that means changing what the sources say, not just the product.

Being described is not being chosen. When the model offers two brands as the standard choice, every other vendor is competing for a lane. Keeper, with a plain business and compliance lane, is picked in 21 of the 25 answers that name it; Dashlane is picked in 15 of 28. Name your lane plainly.

Being built in is not being chosen. Apple Passwords and Google Password Manager ship with Apple devices and Chrome, yet they were picked in 9 answers between them, 5 of those to questions that name an Apple device or the product itself. Default availability does not make you the model's answer.

The limitations of the ChatGPT password manager test start with variance: one run per query means answer variance is unmeasured; Benchmark 01 runs each query three times across three surfaces. The teardowns are collected on different days, so differences between categories mix the category with the date. The brand dictionary covers the general password managers market and deliberately excludes vertical tools, so their appearances are described in prose and not counted. The readers are AI models, not people: they follow a written protocol and agree with each other closely, but a shared blind spot would not show up as disagreement. The line between picked and recommended is a judgment, documented in the protocol with examples. Sentiment is not coded. No vendor-tool cross-check was read for this category. Google's control counts a brand as ranking only when its own domain is in the top ten; a listicle that features the brand does not count. One run per question, one day, United States, English, logged out. Answers vary between runs, so these are frequencies for this sample. The readers are AI models, not people, and the line between picked and recommended is a judgment call; the third reader decided 15 codes in this teardown. The review-and-media list is the codebook's fixed list, so publishers such as Wired count as other third-party pages here.

Every number above can be recomputed from the ChatGPT password manager dataset below, so download it and check it rather than believe it. CC BY 4.0: use the files, cite the page.

• queries.csv: the 100 queries with intent labels.
• brands.csv: the 35-brand dictionary with aliases and canonical domains.
• mentions.csv: 346 coded brand mentions with position, type, a 0/1 picked column and whether the brand's domain was in Google's top ten.
• citations.csv: 292 citation events with domain class and Google overlap.
• observations.csv: one row per query with brand, recommendation, citation and control counts.
• google-control.csv: the Google top ten used as the control for each query, and for the 29 queries whose September 26 page came back swapped, that page beside the October 23 page that replaced it (added October 23, 2026).
• coder-notes.md: coding notes: reader method and agreement, judgment calls that matter in this category, reader-added mentions and dictionary notes.
• codebook.md: the rulebook, with dated amendments through v2.0.
• reader-protocol.md: the written protocol both readers coded against (codebook v2.0).
• picked_stats.json: the picked-level figures.

The earlier teardowns: Teardown 01, project management, Teardown 02, CRM, Teardown 03, email marketing, Teardown 04, help desk, Teardown 05, accounting, Teardown 06, payment processing, Teardown 07, payroll, Teardown 08, HR software, Teardown 09, applicant tracking.

The ChatGPT test we ran on password managers can be run on your own category too. In the wider series: ten categories, and the reading list is still the story.

In password managers the model's reading list is mostly the vendors' own websites, 69 percent of citations, the highest share of the first ten categories, and the brand whose site is cited most is the one it picks most. Finding out what the model is reading for your category is the first thing a Category Salience Brief does, with a query set you approve first.

Moving a brand from named to picked in answers like these is the point of answer engine optimization services.

Marketing teams selling password managers use data like this to plan SEO and AI search for SaaS companies.