News · Explainer
Why do made-up names produce spikes in Google Trends?
AFNPT ran about 1,600 control searches, including invented and cartoon names. Many spiked. The reason is simple: on a term nobody searches, the sample and the noise have nothing real to compete with.
By Americans for Transparency · Published
The short answer
Because the index is relative and sampled. On a term with essentially no real searches, there is no genuine signal, so the sample plus Google's added noise is the entire chart. That produces a flat line with occasional lone spikes to 100. AFNPT tested this with about 1,600 control searches, including invented and cartoon names, and found roughly one rare phrase in three spiking somewhere in the window.
- On a term with no real search volume, the chart is built almost entirely from sampling and noise.
- AFNPT ran about 1,600 control searches to test this, including names no one would search.
- About one rare phrase in three showed a spike somewhere in the window.
- A spike on a made-up name cannot mean someone searched it, so a spike alone proves nothing about real names either.
The logic of the control test
If a method flags real searches, it should stay flat on terms that no one searches. A made-up name is the cleanest possible control: there is no real interest to detect. If the method spikes on it anyway, the spike is coming from the tool, not from the world.[1]
That is exactly what AFNPT set out to check. Using the same settings as the Fort Huachuca searches, it ran about 1,600 control searches, including invented names and cartoon characters, and recorded how often they spiked.[1]
What the controls did
They spiked. About one rare phrase in three showed a spike somewhere in the window, and the dates shifted when the same search was run again. Names that could not possibly have been searched behaved just like the names the posts treated as suspicious.[1]
Why this happens
On a term with no real volume, Google's sample has almost nothing to draw from, and the privacy noise it adds is no longer small by comparison. The 0-to-100 scaling then stretches whatever is there, so a single sampled point or a burst of noise becomes a spike to 100. There is no real signal underneath to anchor the line at zero.[2]
This is why researchers insist on aggregating many samples: averaging many pulls cancels the noise and reveals whether any real interest exists. One chart of one rare term cannot do that.[3]
Questions people ask
If made-up names spike, does that disprove every spike on a real name?
It shows a spike alone proves nothing. A real name might have genuine interest behind it, but you cannot tell from a single rare-term chart, because noise and sampling produce the same picture.
How many control searches were run?
About 1,600, using the same settings as the original Fort Huachuca searches. Roughly one rare phrase in three spiked somewhere in the window.
Why use cartoon characters as controls?
Because no one is plausibly searching them in that region, so any spike must come from the tool. They are a clean test of whether the method flags real activity or just artefacts.
Does this mean Google Trends is useless?
No. It is useful for popular terms and for broad trends measured over many samples. It is unreliable for single charts of rare terms, which is exactly the case the Kirk posts relied on.
Sources
- [1]Americans for Transparency, The Fort Huachuca guest list, checked, October 7, 2026. AFNPT review
- [2]Google, FAQ about Google Trends data. Official source“These apparent one-off spikes should not be interpreted as real search activity or evidence of individual searches.”
- [3]The Conversation (Jacques Raubenheimer, University of Sydney), How misreading Google Trends is fuelling Bondi attack conspiracy theories, December 17, 2025. News report“To get accurate results, one has to aggregate many samples of Google Trends data”