How the Petal Score Works

Our Read, Not a Rating

The Petal Score is our read of a shop's public record, not a measure of how much we like it. It sits on a ten-point scale and lands in one of three bands: Exceptional, Excellent, or Good. The score is always calculated from the five categories below; nobody sets it by hand.

It is not a star rating. We do not mark it up as a rating for Google, and we do not display Google's stars or logos as our own. The shop's Google rating is in the card's Score Details, opened from the score tag, and shows two ways: all time, and the last 24 months, because the recent record is the one that matters for delivery.

National Services: Petal Tested

National flower delivery services are not scored this way. They are judged by Petal Tested, our hands-on test scored out of 5, from orders we place and pay for ourselves, along with the delivered price and tested codes. Those reviews live under our review guidelines.

Not Part of the Score

Three things on our city pages are not part of the score. The typical price is the median of five or more of a shop's own listed arrangements before delivery, leaving out plants, fruit, gift baskets, funeral-service pieces and call-for-price items. It is dated and re-read every quarter, and a shop with no prices online shows none. Clear Prices and Terms records whether prices are published, never how high. A shop's delivery windows, the hours or slots it says it delivers in, are shown on its card from its own website and are not scored. Quick Picks, bouquets from listed shops at their own prices, do not change a score and no shop can ask to be in them.

From Customers

From customers, we measure three things.

What Customers Say is the shop's Google rating over its recent reviews (the last 24 months in full, the two years before at half weight), adjusted so a small number of reviews is pulled toward the city's typical rating; if the last 12 months, with at least 20 reviews, read clearly lower than the full period, the category is marked down.

Delivery Reliability counts how often reviews describe a delivery problem (late or missed, the wrong item or not as pictured, or poor condition on arrival) as a rate across the shop's recent reviews (weighted the same way) and pulled toward the city's typical rate, so one unhappy customer barely moves it. We do not judge individual complaints, only patterns.

When Things Go Wrong records how often the owner answers a critical review (1 to 3 stars) with a reply that engages with the specific complaint rather than a template. It records what the shop did in public, never who was right. The card shows the facts, for example, "Answered 8 of 10 critical reviews in detail."

From Our Checks

From our checks, two measures.

Recognition collects named publication picks, including local press and The Knot's "Best of Weddings" award, and recommendations in local Reddit threads, each read by a person.

Clear Prices and Terms records whether the shop's own website publishes prices, a same-day cut-off, a delivery area and a delivery fee. A website we cannot read is treated as a gap in our data, never a mark against the shop.

Both checks start at 6 when nothing is found and reach full marks (10) with enough evidence. For Recognition, three points reach full marks: a recent named publication pick adds about 1.3; an older pick, one or two Reddit threads or a recent Knot win adds about 0.7. Neither check decides whether a shop is listed, so a shop that does not market itself stays listed if its customer and delivery record hold up, and it can still reach the top bands.

Beyond the public record, we may contact individual florists to validate details and data points, visit shops in person, and place anonymous orders of our own. These checks corroborate facts and the patterns in customer reviews and confirm what a shop offers.

One Score, Always Calculated

The overall score is the average of the five, to one decimal, always calculated and never set by hand.

Where the Reviews Come From

Google reviews come to us through a commercial data provider. Reviewer names, photos and profile links are removed on import. The review text stays private; we publish only counts and figures, each with its source and date.

What the AI Does

One AI model has one job: it reads each review and sorts it into themes such as "late or missed." It is trusted only after it matched people. On 50 reviews that our founder, James Sinclair, sorted without seeing the model's answers, he and the model agreed on whether a review described a delivery problem at a level statisticians accept as strong agreement. AI does not score shops or decide who is listed. This is the same AI work described in our review guidelines. This page's words, like every editorial page, may be drafted with an AI tool before James Sinclair, our founder, reviews, edits and signs it off with a date.

What People Do

People do the rest: a person checks every listed shop against the seven rules, and the card shows the date. People also do the authenticity read of every shop with a warning sign, the Reddit read, a written read of every shop listed on thin evidence or cleared from review, and any shop sent for review.

If a shop asked walk-in customers for reviews over a short period, those reviews count at half weight, and a person spots these in the authenticity read. Every refresh also checks that each shop's reviews came back in full against Google's own count and the last read. A shop whose reviews come back thin is not rescored from that read until a person looks.