Search evidence · Reading relative interest

What do Google Trends numbers mean?

A Google Trends score of 100 is a reference point inside a selected comparison, not a count of searches. Before repeating a claim from a Trends screenshot, recover its search definition, place, dates and filters. Without those, the line has lost the information that gives it meaning.

This guide concerns the relative charts in Trends Explore. It gives you a small calculation exercise and a way to turn a screenshot into a finding another researcher can check.

Two operations sit behind the scale

First, the search interest is considered relative to all searches in the relevant place and period. Then those proportions are indexed against a peak. Google’s Trends fundamentals lesson explains normalization and indexing separately. Changing the comparison can therefore change the displayed numbers without rewriting what happened.

Imagine a fictional search dataset for “repair café.” The counts below are invented teaching inputs, not data downloaded from Google. We make the denominator visible because an ordinary Trends chart does not hand you these raw counts.

Our simplified calculation is share = selected searches ÷ all searches, followed by index = share ÷ highest share × 100. It illustrates the scale, not Google’s full sampling and processing system.

From A to B, the raw count rises from 100 to 150 while the index falls from 50 to 25. “Interest fell by half” needs the word relative and the chosen comparison attached. “Half as many people searched” is unsupported twice over: the count rose in this exercise, and searches are not a count of distinct people.

Now keep only weeks A and B. The highest share is 1%, so their new indices are 100 and 50. Week A has changed from 50 to 100 purely because the reference window changed. Put the two charts side by side without their dates and a reader could mistake rescaling for growth.

Choose the thing you mean to measure

A term and a topic are different inputs. A term follows the chosen wording; a topic groups terms expressing a concept across languages. Google’s comparison guidance explains that quoted terms, unquoted words and topics can cover different searches.

For example, “jaguar” could describe an animal, a car-related search or another use. Write the question before choosing the input: “Interest in the animal” needs a different definition from “searches containing this exact spelling.” Neither choice is universally better. The problem is silently switching definitions halfway through a comparison.

For a bilingual investigation, save the exact Persian and English terms, including spacing, and state whether you grouped them. A translated word does not automatically represent the same collection of searches. If you use an available topic, record its displayed topic label rather than presenting it as an exact-word query.

Compare together before comparing screenshots

Suppose two independently exported charts each reach 100. That tells you both have a peak within their own scale; it does not establish equal search demand. Rebuild the intended comparison with the same place, dates, category and search type, and put the intended terms together. Google’s data interpretation lesson shows why a relative line is not an absolute volume series.

A useful check is to change one setting deliberately and record the effect. Compare one month with a longer window, keeping everything else fixed. Then restore the original period before testing a different term. If you change three controls at once, you will not know which change explains the difference.

Do not attach yesterday’s export to today’s screen and assume their scales agree. Keep each export beside its settings. A repeated lookup is a fresh observation; it should not overwrite the file used to support an earlier published finding.

Read zero, spikes and “Breakout” differently

Google’s data FAQ says the service uses sampled search data, low-volume terms may appear as zero, and statistical noise can create apparent spikes, especially at very low interest. A tiny isolated spike cannot identify an individual searcher or establish their intentions. Search interest is also not a scientific opinion poll.

Use zero as the displayed result under the recorded settings. Do not convert it into “nobody searched.” If a claim depends on a small spike, test whether it survives a sensible wider context and seek a separate source for the alleged event. A news report, event record or dated publication can address a different part of the story; another cropped chart cannot supply the missing cause.

“Breakout” belongs to rising related searches. Google defines it as growth above 5,000% in that comparison, not a search total. See related-search definitions. In a separate arithmetic example, 10 becoming 610 is 6,000% growth; 10,000 becoming 15,000 is 50%. The larger percentage does not identify the larger audience. These invented counts do not reconstruct Google’s hidden inputs.

Send a chart with its reading instructions

Google documents CSV export and source attribution. Save the downloadable chart data when available, the comparison link and an image that includes the controls. Credit Google Trends when you reuse its data. Keep the untouched download separately from calculations you make afterward.

Use this compact record for the handoff:

• Question: the specific comparison the chart is meant to answer.
• Inputs: exact terms or topic labels; any grouping.
• Scope: place, date range, category, search type and displayed time zone where relevant.
• Observation: capture time, source link, original export and values actually shown.
• Reading: one sentence describing the relative result, plus what it cannot establish.

For our exercise, a defensible finding is: “The selected search share was lower in B than A; the available raw teaching counts nevertheless rose.” For a real screenshot without a recoverable definition, write “comparison settings unavailable” before attempting a growth claim.

If this chart appears in a larger investigation, use OSINT Jet’s guide to reading an OSINT report to check whether the conclusion stays within the evidence. The useful deliverable is a reproducible comparison and a bounded statement, not a dramatic line with an unexplained percentage.