How Google Trends Is Used to Measure Academic Cheating

TL;DR

Google Trends provides country-level search volume signal for keyword queries. The AMI uses this for D1 (contract cheating) and D2 (AI submission) dimensions. Strengths: live data, broad country coverage, language-specific signal. Limitations: measures demand not confirmed incidence, signal includes non-student discussion (Norway anomaly), language markets complicate cross-country comparison.

What Google Trends provides

Google Trends is a public Google service that reports the relative search volume for specific queries across geographies and time. The data is:

• Relative, not absolute — Google does not publish raw search counts; values are normalised
• Country-level (and sometimes sub-country) — queries can be run at country, region, or city resolution
• Time-windowed — historical data available back to 2004, with recent data available daily
• Language-specific — different language variants can be queried separately

For the AMI, the API returns a 0–100 score per country for a given query and time window, with 100 representing the country with the highest signal.

How the AMI uses Google Trends

D1 Contract cheating

The AMI runs two sets of D1 queries:

Generic contract cheating terms (English and other major languages):

• "buy essay online"
• "essay writing service"
• "pay someone to write essay"
• "do my assignment"
• equivalents in Spanish, Portuguese, French, Italian, Polish, etc.

Essay mill brand names:

• "ukessays"
• "edubirdie"
• "papersowl"
• "easyessay"
• regional equivalents

Brand name queries are more precise signals — generic terms can capture educator discussion ("how can teachers detect essay mills?"), but brand name queries reflect actual demand intent.

D2 AI submissions

The AMI runs queries for AI submission tools:

• "chatgpt for essays"
• "ai essay writer"
• "claude for homework"
• equivalent terms in major languages
• AI bypass tool brand names

Normalisation

The 2022–2026 4-year window captures the post-ChatGPT period (ChatGPT launched November 2022). The country with the highest signal across this window scores 100; others scale relative.

Limitations

Demand not incidence

Google Trends measures *searches* for terms — not confirmed cases of misconduct. A student searching "buy essay online" may not actually purchase; an educator searching to write a policy may not be planning misconduct. The signal correlates with demand but does not directly measure incidence.

Academic and policy discussion contribution

This is the Norway anomaly. Countries with high digital engagement and active academic and policy discussion of integrity topics produce search volume that is not student demand. Researchers writing about AI submissions, journalists reporting on essay mills, policymakers drafting legislation — all contribute to the same search signal that captures student demand.

The countries most affected are typically those with strong open academic discussion: Nordic countries, Netherlands, Germany, parts of the UK. The AMI methodology flags Norway as the principal case and notes that Sweden shows a milder version of the same pattern.

Language markets

Google Trends signals are language-specific. A query in English captures English-language search; the same query in Spanish captures Spanish-language search.

For some countries, all major searches happen in the country's principal language — there is no spillover problem. For others (e.g. small Anglophone populations in non-English countries), language markets cross country boundaries. Spanish queries for essay mills can come from any Spanish-speaking country; Spanish countries with high search volume may be partly capturing demand from other Spanish-speaking countries.

The AMI methodology accounts for this through language-disambiguated query batches but the approach is imperfect.

Small countries with low absolute volume

Countries with small populations may have low absolute search volume, producing noisier per-capita signal estimates. Singapore, Ireland, New Zealand show some of this effect.

Brand name decay

Essay mill brand names change over time as companies rebrand, get sued, or move jurisdictions. Brand name queries become less reliable over multi-year windows as the underlying brands shift.

What the data shows

D1 distribution from the v1.5 dataset:

• Top: Colombia (100), Argentina (100), Greece (100), Pakistan (100) — multiple tied
• Middle: most European countries (50–67)
• Bottom: Australia (33), UK (33), Ireland (33), Canada (50)

The lowest D1 scores in the dataset are in countries with specific contract cheating bans. This is consistent with both the legislation reducing actual demand and reducing the brand name search volume for major essay mill services.

Future improvements

The AMI methodology document discusses planned improvements:

• Language-disambiguated query batches with cross-country attribution
• Weighting non-student search-source contribution out of the signal
• Combination with confirmed-case data where available (FOI in UK, ORI in US)
• Sub-national variation analysis where Google Trends supports it

Sources

• Google Trends API documentation
• AMI v1.5 methodology document
• Methodology caveat section on Norway anomaly

Full methodology | Download dataset

Related

Read the full methodology