Three young companies own the machines’ memory.

Who the machines recommend

1Finding 1: The new guard owns the machines’ memory

Deel, Rippling and Remote take more brand mentions than the rest of the category combined. The pattern repeats what our CRM report found with Microsoft Dynamics, now at category scale: revenue rank and machine rank are different leaderboards. AI assistants learned this category from the internet of the last five years, and that internet talks about remote work, global hiring and the tools built for both. Every incumbent strategy deck that assumes brand equity carries over into AI answers should be tested against this table.

2Finding 2: Workday is the big-company answer

Workday appears in effectively every answer about enterprise HR platforms and in 90 percent of answers naming the best platform overall. Outside those two questions it thins out fast: 38 percent for a startup’s HR stack, around 20 percent for a first HR system and for payroll across countries, and zero on hiring abroad and paying contractors. That is either a razor-sharp position or a slow leak of tomorrow’s mid-market pipeline, and only Workday can decide which. What the numbers rule out is the comfortable assumption that a large enough brand is automatically in the conversation. It is not, and the mid-market buyers it might want next are the ones asking the questions where it is weakest.

3Finding 3: Europe’s champion is invisible to the machines

Personio serves thousands of European mid-size companies. In the machines’ answers it gets 1 percent of brand mentions from memory, covers three situations of eight, and search lifts it only to 2 percent, which means there is no content operation compensating for the memory gap either. Every day, European companies ask assistants about HR software in English and get an American and global-remote shortlist. Whether the same holds in German is a question this report deliberately leaves open; measuring it is a two-week exercise, and we suspect the answer is worth more to Personio’s board than most of its current dashboard spend.

4Finding 4: Eight situations, three markets

The category splits into three machine-markets that rarely mix: global work belongs to Deel and Remote, domestic and startup questions to Rippling and Gusto, enterprise to Workday alone. A brand’s growth plan here is not one battle. It is a choice of which of these situations to enter, and the table shows the incumbent you would be entering against.

5Finding 5: The answers are fed by the winners’ own websites

When assistants search before answering, they cite sources, and the top of the list is not the HR press: deel.com (87 citations), rippling.com (73), remote.com (52), gusto.com (48). What the leaders publish is being read back to buyers as the answer. For everyone else this is the actionable part: in this category the machines demonstrably read vendor content, so the pages that answer buyers’ questions are a seat at the table, and the full citation list, which we hold for every brand in this report, is the target list for earning one.

A recorded answer, in context

One answer, verbatim, from the archive behind this report. The study’s results come from repeated questions; a single answer shows the format buyers see.

“We want to hire an employee in another country, what platform handles that?”

Recorded test question

To hire a full-time employee in another country without setting up your own local legal entity, you need an Employer of Record (EOR) platform.

• Deel: best for speed, ease of use, and overall versatility.
• Remote: best for intellectual property (IP) protection and stability.
• Rippling: best for companies that want HR, payroll, and IT in one single system.

Gemini via the Google API, from memory, August 2026. Opening excerpt, not the entire response.

What we would do about it

A taste of the advisory layer. Each of these is a decision with a number attached, not a tactic from a listicle:

• If you lead here: your risk is not the rival next to you in the table, it is the situations you have not entered. Rippling at 30 percent in enterprise against a lone Workday is the most interesting open flank in the category.
• If you are Workday-shaped, a giant that is thin on most of the machines’ shortlists: first decide whether that is strategy or neglect. If it is neglect, the fix starts with entity work and presence in the specific sources that feed the situations you want, and it is measured in quarters. Anyone promising it in weeks is lying to you.
• If you are Personio-shaped, a regional champion invisible in global answers: measure your home language before assuming it is safe, then build from your stronghold situation outward. Three situations at high strength beat eight at 1 percent.
• If you are anyone else: pick the one situation where you have a real claim and go from present to default there. The table above tells you exactly how crowded each door is.

The paid version is specific to you: your numbers per situation, the false claims AI makes about you, the exact sources to earn, priorities with owners and effort grades, and the tactics to skip. That is the audit, and the Program that carries the work out. To see the full depth on a real brand, read the sample audit of monday CRM built from our CRM dataset.

How we measured

ChatGPT, Gemini and Claude, queried through their official developer interfaces in August 2026, in the model tiers most buyers actually use. 1,008 recorded answers across eight buying situations, in two modes: from memory and with web search, reported separately. The design was frozen and hashed before the first question, and this category is the most stable we have measured (split-half r = 0.99); differences under 15 points per situation are noise and we treat them as such. These are controlled tests, not private customer conversations, and results can differ from the consumer apps.

Definitions, limits and the pre-registration protocol: methodology v0.2. The brands measured did not participate in, commission, or endorse this report, and we never sell placement in AI answers. Commissioned audits never change published scores.