How the numbers are made, and where they come from

01How a finder report is made today

1. Your description is read with a dictionary of words and patterns, to work out what kind of business you run and the words you use for your customers and what you sell.
2. Your answers pick audience patterns from a library written for different kinds of businesses, and each pattern is scored against your answers.
3. Your goal reweights those scores: which audience traits count most, which channels rise or fall, how the budget leans, which actions are eligible, and the order of the sections.
4. Channels are ranked for each audience, and the budget is split across them from their scores and the range you gave.
5. The report is compared with one built from the same answers and no goal, and the differences are listed in "What your goal changed".

• No AI model is involved. The rules are written by people.
• The same answers always give the same report.
• The report is built on the uraudience server from your answers, and saved there with them at its own address. Without an account, only the browser that made it can open it. With an account, you can open it on any device you sign in on.
• Shares and scores in a finder report come from these rules, not from counts of real people, and are labeled as estimates with a confidence mark.

04The tables behind a count

People and household counts come from the American Community Survey 5-year Summary File, stored for the nation, each state, every Census place, and every ZIP code area, with the margin of error for each figure. Business counts come from County Business Patterns for states and counties, and ZIP Code Business Patterns for ZIP codes. At the ZIP code level the Census Bureau withholds many counts by business size, so that no single business can be picked out.

05The assumptions we state

Every estimate rests on one or more of these, and a figure’s details list the ones it uses, in these words.

• Where two filters come from different Census tables, we assume they are unrelated inside each small area, then add the areas up. Real people are not always that independent, so the result is shown as a range.
• For whole states or the nation, we assume the filters are unrelated across the whole area. That loses local patterns, so confidence drops one tick.
• Household income is published for households, not people. We apply each area's household income mix to its adults.
• The Census publishes this table in broad age bands, so your age range was widened to the nearest published bands.
• Counted by ZIP Code Tabulation Area (ZCTA), the Census Bureau's version of a ZIP code area. Most ZIP codes have one. PO box and single-building ZIP codes do not.
• A ZIP code area is included when its Census center point (the internal point) falls inside the radius.
• The margin of error for a sum of Census cells is estimated as the square root of the sum of the squared cell margins, the approximation the Census Bureau's ACS handbook describes.
• Some Census cells in this area are controlled estimates: the Census Bureau fixes them to its official population estimates and publishes no margin of error for them. They add nothing to the margin of error shown.
• Two tables count slightly different groups of people, for example employed people 16 and over and adults 18 and over, so combining them is approximate.
• Interest and behavior rates are uraudience estimates, not measurements, and are deliberately wide.
• This interest rate comes from a published survey and is applied as if it were the same everywhere.
• A job title's share of its occupation group comes from BLS figures for the whole state, applied to smaller areas.
• BLS counts jobs where people work; the Census counts people where they live.
• People are counted through the occupations where O*NET rates the skill as important, not through their own skills.
• County Business Patterns counts business locations (establishments), not companies. A chain with forty stores counts forty times, and businesses with no employees are not included.
• A city is counted through the ZIP code areas that approximate it, so the edges do not match the city line exactly.
• Cost per thousand views is a uraudience benchmark range, not a quote. Ad platforms set prices by auction.
• Assumes each person sees your ad a set number of times a month, shown next to the estimate.
• Assumes only part of the Census audience can be reached on each platform. The share is a uraudience estimate.
• Finder audiences come from uraudience rules matched to your description. Their shares are estimates, not measurements.
• BLS publishes a relative standard error for each job count. The margin shown is 1.645 times the standard error, the same 90 percent level the Census Bureau uses, added across states and job titles as the square root of the sum of squares.
• ZIP Code Business Patterns does not publish every size class for every area. Where one is missing, the area’s published count for that industry is split across the missing size classes in the proportions its state publishes. That part is an estimate, shown as a range of half of it either way.

06Words we use carefully

Sex

The Census asks about sex, not gender identity, so the filter is called Sex. Ad platforms use their own gender settings.

ZIP code areas (ZCTAs)

ZIP Code Tabulation Areas, the Census Bureau’s version of ZIP code areas. Most ZIP codes have one. PO box and single-building ZIP codes do not.

City

A Census place: a city, town, village, or census-designated place, shown with the Census name, such as "Evanston city".

Business locations

Establishments, not companies. One company with three shops is three business locations.

Center point

The Census internal point of a ZIP code area. A radius includes an area when its center point is inside the radius.

08Export formats, and when we last checked them

A file for an ad platform ships only where that platform's own documentation confirms its format. Everything else goes on the platform's spec sheet, which lists the settings to enter by hand. Every format below was last checked against the linked page on September 11, 2026.