We Asked Google's AI How It Picks Cleaning Companies. It Answered With Percentages.

In late August we ran an experiment. We took the fifty questions office and facilities managers actually ask before hiring a commercial cleaning company — the same fifty we answer on this site — and put each one into Google, capturing what the AI Overview said and which sources it drew on. Then we asked Google's AI Mode a question nobody usually gets to ask a search engine directly:

When you recommend specific commercial office cleaning companies in Manhattan, based on which factors exactly do you choose them? Rank the factors by how much weight each one carries.

It answered with a numbered list and percentages. We are publishing it in full, because it is the most honest procurement checklist we have seen, and because it did not come from us.

What Did Google Say the Factors Were?

Treat the exact percentages as indicative rather than as a published algorithm — a language model producing a number is not the same as a documented ranking system, and a different phrasing on a different day would produce slightly different figures. What is durable is the order, and the order matches what experienced facilities managers already do.

Why Is Insurance Weighted so Much Higher Than Everything Else?

Because in a managed Manhattan building it is not a preference, it is a gate. Google's answer named the specific mechanics: a Certificate of Insurance issued for your particular property, general liability and umbrella coverage at the limits the landlord demands, bonding against theft, workers' compensation, and compliance with state labor law and, where applicable, 32BJ SEIU agreements.

A vendor who cannot produce that paperwork does not get a poor score. They get turned away at the freight elevator, and you find out on the first night. Everything else on the list is a question of quality; this one is a question of whether the work can happen at all.

What Does “Operational Flexibility” Mean At 20%?

Google broke this into three things: after-hours and weekend coverage, including overnight windows around 8pm to 4am; the ability to deploy a crew at short notice for a leak, an event or a sudden office change; and scalability from a small suite to a multi-floor account.

This is the factor most cleaning companies are quietly weak on and never mention on their website, because it is easier to publish a phone number than to publish what happens when you call it at 3am. Ours is on a page of its own.

Why Does Price Carry Only 5%?

This is the single most quotable line in the whole answer. Google put price last and explained why: cutting corners on price in Manhattan usually results in a failure of the first or second factor, leading to building lockouts or poor service.

That is not an argument for paying more. It is an argument for knowing what you are buying. A lower price is only meaningful if the scope, the hours and the insurance are identical — and when a bid comes in well under the others on an identical written scope, the variable that absorbed the difference is almost always labor hours. Why two quotes for the same office differ by 40%.

Worth noting what else Google said under this heading: it prefers itemized bids over vague flat rates, and month-to-month or reasonable cancellation terms over restrictive multi-year lock-ins. If your current vendor's proposal is a single number and a three-year term, that is two strikes on the factor they are probably competing hardest on.

What Came up Most Across All 48 AI Overviews?

Separately from the ranked answer, we counted which concepts appeared across the AI Overview text for all fifty questions. The frequency tells you what the model reaches for when it describes this industry at all:

Paper first, people second, price third. The two certifications that keep surfacing — Green Seal and ISSA CIMS — are worth understanding before you let a logo do any work in your decision, because the most common misuse is a company using certified products describing itself as a certified company. What each one actually proves, and which ones we hold.

What Happens When You Search for Companies Rather Than for Answers?

Something different, and it is the most useful practical finding of the whole exercise. Questions like which cleaning companies serve Hudson Yards or which janitorial companies work in Midtown East returned no AI Overview at all. Google answers those with the local map pack — the three businesses on a map, with star ratings.

For a buyer, that is worth knowing: the shortlist you are shown for a location search is driven by proximity, review volume and Google Business Profile completeness, not by how good the companies' websites are. If you are choosing a vendor from that map, you are choosing from a list ranked partly by marketing diligence. Do the four-minute verification on whoever appears there rather than assuming the order means something.

Does Any of This Change How You Should Choose a Vendor?

Honestly, no — and that is the reassuring part. The ranked list is not a strange machine preference. It is roughly the order a careful facilities manager would put those factors in anyway: can they legally and practically get into my building, can they come when I need them, who are the people, do other buildings vouch for them, can they do the specific work, and only then, what does it cost.

What it does change is how much weight to give a cheap bid. If the thing that decides your outcome is 35% insurance and 20% availability, and price is 5%, then a vendor who leads with price is competing on the least predictive variable in the set.

Method: fifty buyer questions run as individual Google searches (en-US) on 26 August 2026; AI Overview text and outbound source domains captured for each; the ranking question put separately to Google AI Mode. Forty-eight returned an AI Overview, two returned a map pack. AI Overviews vary by user, location and session, so treat this as a snapshot rather than a constant.