AI Engines Mention Bitwarden Most for Enterprise Password Managers. Four of Six Name 1Password First.

An IT director with 400 staff opens an AI engine and types the question every password-manager vendor is trying to be the answer to: which enterprise password manager should we buy? The engine returns three to six names, a sentence on SSO and SCIM for each, and a recommendation. There is no page two. Whichever vendor is named first has just been shortlisted, and whichever vendor is absent has just been excluded, before anyone at that company has visited a vendor site.

We wanted to know what those engines actually say in this category, measured across every major engine at once rather than inferred from a handful of prompts on one of them.

Key Takeaways

• Bitwarden led with 235,000 mentions, Keeper followed at 218,000 and 1Password at 214,000. The three took 61.5% of all mentions across the 33 brand strings recorded.
• 1Password was named first in four of six engines. Bitwarden opened the response only in Google AI Mode and Google AI Overview. Keeper, second on mentions, opened none and was read later than both in every engine.
• The engines' cited domains overlapped by a mean Jaccard coefficient of 0.14, yet the same vendor site was the most-cited domain for five of the six.
• 58.9% of all cited pages sit on a vendor's own site. ChatGPT draws 78.8% of its citations from vendor domains; Microsoft Copilot draws 9.9%.
• Only 0.9% of cited pages were dead, and the same three brands led all four markets in the same order.

How We Ran the Study

We issued 10,000 enterprise buyer-intent queries about password managers, credential vaulting and workforce secrets to six AI engines: ChatGPT, Perplexity, Gemini, Microsoft Copilot, Google AI Overview and Google AI Mode. Every query ran in the United States, Canada, India and Germany during August 2026, giving 40,000 responses per engine and 240,000 in total.

For each response we recorded its length, every page it cited and what kind of site that page sits on, whether the page still resolved, every brand it named, and the ordinal position at which each brand first appeared. Spelling variants of one brand and plan variants of one product were merged before counting. The corpus held 4,592,000 cited pages, 31 distinct product entries and 33 brand strings. The counting method behind those figures is the one we set out in how to measure AI share of voice across engines.

Two limits apply. The engines are commercial services with no version pinning, so a repeat run would not return identical text. The query set is purposive rather than random. We report descriptive statistics only and make no significance claims. GrackerAI sells AI search visibility measurement for categories including this one; no vendor paid for, requested or reviewed any part of this study. The full method, every table and the validity discussion are in the complete benchmark report.

Bitwarden Is Mentioned Most. 1Password Is Named First. Keeper Is Neither.

Three measurements, three different leaders.

Bitwarden recorded 235,000 mentions across the corpus, the highest of any brand, and Bitwarden's own site was the most-cited domain in five of six engines. Yet when we asked which brand most often opened the response, four engines said 1Password: ChatGPT, Microsoft Copilot, Gemini and Perplexity. Only Google AI Mode and Google AI Overview opened with Bitwarden.

Keeper is the third profile. It recorded 218,000 mentions, 4,000 more than 1Password, and it opened the response in no engine at all. Its mean ordinal position ran from 2.3 to 2.9, later than both Bitwarden and 1Password in every one of the six.

1Password recorded the earlier average position in four engines. The two brands tied at 1.8 in Gemini. Bitwarden was earlier only in Google AI Overview, at 1.5 against 1.7. The widest gap was in Perplexity, where 1Password averaged 1.2, the earliest brand-level mean position recorded in the study, and Bitwarden averaged 2.0.

Our finding: Google AI Mode disagrees with itself. It opened with Bitwarden more often than with any other brand, yet 1Password recorded the earlier mean position there, 1.7 against 1.9. A brand can win the first slot most often and still average later, if it is placed first in a majority of responses and much further down in the rest.

That third number matters because a dashboard that reports only mention count, or only average position, misses it. Keeper shows why: on mentions it is second, 4,000 ahead of 1Password, on opening position it is nowhere, and a vendor in that position has a different problem from one that leads on mentions and opens second.

Three Brands Take 61.5% of All Brand Mentions

The category is concentrated at the top and fragmented everywhere else.

Bitwarden, Keeper and 1Password together took 61.5% of the mentions recorded across the 33 brand strings. Dashlane followed at 139,000 and NordPass at 87,000; below NordPass no brand exceeded 29,000. The gap between the third and fourth brands, 75,000 mentions, is larger than the spread across the whole leading three. That kind of settled core is not unique to this category; a separate analysis found AI engines picking the same three DSPM vendors, though that study ran seven engines rather than six.

Seven of the 33 brands were named by all six engines: Bitwarden, Keeper, 1Password, Dashlane, NordPass, Psono and Passbolt. Fourteen were named by exactly one engine, none of them above 4,000 mentions, and the list of those fourteen shows where the engines drift out of the category: identity services, secrets managers, consumer credential stores and, in one case, an analyst firm. Of the 31 product entries, 7 were named by all six engines and 14 by exactly one.

The Engines Agree on One Domain and Little Else

The six engines reached a similar shortlist from largely separate reading lists, with one exception.

We measured the overlap between each pair of engines' cited domains as a Jaccard coefficient, where 1 means identical sets and 0 means nothing in common. The mean across all 15 pairs was 0.14. Only three pairs reached 0.20. The highest value, 0.27, was between Gemini and Perplexity, two engines with different parents; the two Google search surfaces scored 0.26 with each other. Microsoft Copilot recorded the lowest coefficient with every other engine, from 0.03 with Google AI Mode to 0.08 with Perplexity, and its most-cited domain appears in no other engine's top three.

The exception is a single site. Bitwarden's own website was the most-cited domain for ChatGPT, Gemini, Google AI Mode, Google AI Overview and Perplexity. The engines read different parts of the web, but the one page they all read most belongs to the vendor they mention most. We recorded citation and recommendation from the same response, so this does not tell us which caused which.

Volume compounds the split. ChatGPT cited 2,730,000 pages, 59.5% of the entire corpus. Google AI Mode cited from the widest population of domains at 167, ChatGPT from 124, and Perplexity from the narrowest at 32. A source list that earns citations in ChatGPT describes a small part of what Microsoft Copilot reads. Why the engines diverge this far starts with how each one chooses what to read, which we set out in how the major AI engines decide which sources to cite.

Where the Evidence Comes From: Vendor Sites

More than half of everything the engines cite is the vendors talking about themselves.

Across the corpus, 58.9% of cited pages sit on a domain belonging to a vendor named in the response, 2,706,000 of 4,592,000. The share divides the engines. ChatGPT draws 78.8% of its citations from vendor sites, and since it produces 59.5% of all citations, it sets the corpus-wide figure almost by itself. Gemini draws 42.2%. The two Google search surfaces and Perplexity sit in the middle, with review aggregators, forums, video and news making up a substantial minority of what they cite; review sites reach 16.2% of Perplexity's citations and 15.0% of Gemini's.

Microsoft Copilot is the opposite case. It draws 9.9% of its citations from vendor sites and 77.8% from domains outside every named class, led by a statistics aggregator. Six engines arrive at the same three vendors from evidence mixes that range from almost entirely vendor documentation to almost none of it.

For a vendor, this is the clearest instruction in the study. The engine that produces most of the category's citations reads your own domain. The others add third-party comparison and community pages on top. We classified domains, not pages, so which pages on a vendor's site were cited is an implication rather than a measurement. Finding out which of your pages an engine is citing takes a per-engine workflow, and ours is in how to track AI citations, mentions and sources.

Almost Nothing the Engines Cite Is Dead

Of the 4,592,000 pages the engines cited, 39,190 did not resolve when we tested them during the collection window. That is 0.9%.

No engine exceeded 2.1%, the rate for Perplexity. Gemini recorded 0.0%, ChatGPT and Google AI Mode 0.7%, Google AI Overview 1.1%, Microsoft Copilot 1.2%. Because ChatGPT cites so much, its 0.7% still amounted to 19,100 dead links, 48.7% of all of them. No engine cited its own parent's properties at a measurable rate, and none refused a query.

We counted a page as dead only when it returned not-found, gone or a server error, or when its host could not be reached after a retry. A page that refused our request, as sites behind bot protection do, counts as reachable, because the page exists. On that definition, link rot is not a factor in this category. The pages the engines lean on resolved at collection time.

The Engines Disagree on HIPAA and Self-Hosted SCIM

The disagreements in this category are not about who the vendors are. They are about what the vendors' products can do.

ChatGPT returned the statement that 1Password does not sign HIPAA business associate agreements, attributing this to its zero-knowledge architecture. Gemini sometimes listed 1Password as HIPAA-compliant without that caveat. Microsoft Copilot stated that SCIM provisioning works on a self-hosted Bitwarden Enterprise deployment, while some Gemini responses described it as a beta capability or as needing extra configuration. Gemini alone described Vaultwarden as a lightweight Bitwarden-compatible option that lacks official enterprise SAML and SCIM support.

The tail also drifts out of the category. Microsoft Copilot alone returned IdentSphere, which it described as a self-hosted identity infrastructure alternative rather than a password manager, and it recorded 15,000 of CyberArk's 16,000 mentions. In one passkey query it named Microsoft Entra ID before any password manager. Perplexity alone named Azure Key Vault, AWS Secrets Manager, Descope and JumpCloud, at 1,000 mentions each. When a buyer asks about password managers, some engines answer with secrets managers and identity services.

Nothing Changes by Country

We ran the same 10,000 queries in the United States, Canada, India and Germany. The result is the same in all four.

Bitwarden led every market. Keeper was second in every market. 1Password was third in every market, so the market order is the order of the overall mention totals. No market-specific vendor entered the leading positions anywhere, and 12 of the 31 product entries were recorded in all four markets, including every one of the ten most-mentioned.

This is a stronger result than we found in the categories we measured earlier in this series, where the leading vendors held but their order moved between markets. It is also not universal: our analysis of what AI engines recommend in network security found recommendations diverging by region, though it covered seven engines across ten markets, so the difference may be scope rather than category. Here even the order held. If your brand is missing from the shortlist in one of these four markets, it is missing in all of them, and a localized landing page is not what fixes it.

Six Answer Shapes for One Question

The engines that agree about vendors do not agree about how to answer.

Mean length spans a factor of 5.1, from 97.5 words in Perplexity to 492.9 in ChatGPT. Perplexity also hedged most, in 28% of responses, and cut off mid-sentence most, in 42%. Microsoft Copilot anchored 63% of its responses to an explicit date and never truncated. As an illustration rather than a measurement: a 97-word response has room for three vendor names and a compliance acronym, and a 493-word response has room for a comparison table and a caveat about business associate agreements.

What Vendors Should Do With This

Report three numbers, not one. Mention volume, first-mention share and mean position moved independently here. Keeper is second on the first and absent on the second. If your dashboard shows one figure per vendor, it is hiding the ordering problem. If you are choosing instrumentation, we compared 15 AI search monitoring tools on exactly that.

Treat your own site as the primary citation surface. 58.9% of all citations point at vendor domains, and the engine that produces most of them draws 78.8% of its evidence from there. Documentation, feature matrices and compliance pages are what gets read. Third-party review and community pages are what the other engines add on top.

Write the deployment-specific facts down. Two of the four recorded disagreements turned on whether a specific capability exists on a specific deployment model or under a specific agreement: SCIM on self-hosted, a signed BAA. A page that states which capabilities hold on which deployment gives the engines something to extract and removes the ambiguity that produced the disagreement.

Measure all six engines. With mean source overlap at 0.14 and Microsoft Copilot below 0.10 with everyone, a program tuned to ChatGPT tells you almost nothing about Microsoft Copilot. The manual version is to run every prompt across every engine and log what each one cites.

Skip localization for these four markets. The shortlist and its order were identical in all of them.

Do not budget for link rot here. 0.9% of cited pages were dead. In this category the pages the engines cite are almost all there.

Frequently Asked Questions

Which enterprise password manager do AI engines mention most?

Bitwarden, with 235,000 mentions across 240,000 responses from six AI engines in four markets, ahead of Keeper at 218,000 and 1Password at 214,000.

Which password manager do AI engines name first?

1Password, in four of six engines: ChatGPT, Microsoft Copilot, Gemini and Perplexity. Google AI Mode and Google AI Overview named Bitwarden first most often. Keeper was named first by none.

How many AI responses did this password manager study analyze?

240,000 responses, from 10,000 enterprise buyer-intent queries issued to six AI engines across the United States, Canada, India and Germany in August 2026. The corpus contained 4,592,000 cited pages and 31 distinct product entries.

Do AI engines cite the same sources for password manager recommendations?

Mostly no. Mean cited-domain overlap between engine pairs was 0.14 on a scale where 1 is identical, and only three of 15 pairs reached 0.20. The one shared habit was that five of six engines cited Bitwarden's own website more than any other domain, and 58.9% of all cited pages sit on some vendor's own site.

What share of AI-cited sources about password managers are dead links?

0.9% across this study, ranging from 0.0% in Gemini to 2.1% in Perplexity.

Final Thoughts

The engines agree about the vendors and disagree about the order. Bitwarden is mentioned most and opens the response in the two Google surfaces. 1Password opens it everywhere else. Keeper sits between them on mentions and behind both on position in every engine.

For a vendor in this category, inclusion is the first problem and ordering is the second, and they are separate problems. The engines reach their shortlist from evidence bases that overlap by 0.14, reading different parts of the web and converging on one vendor's site, with the largest of them reading vendor documentation almost exclusively. Entering the shortlist has to be solved once per engine. Opening the response is a different piece of work again.

We ran the identical protocol in an earlier category and found a looser version of the same shape: which attack surface management platforms AI engines recommend, also 240,000 responses across six engines and four markets.

The full methodology, every per-engine table and the validity limits are in the complete benchmark report.