GEO tactic evidence lookup
Search the 14 most-recommended AI search tactics and see how strong the evidence really is, and which study it comes from. Data from the GEO Tactic Evidence Scoreboard.
Source: GEO Tactic Evidence Scoreboard, 14 tactics graded September 2026. Grades are copied unchanged.
4 steps, no signup
- 01
Search a tactic
Type what you were told to do, such as
schema,llms.txtorbacklinks. - 02
Filter by grade
Show only Fact, Evidence, Hypothesis or Open tactics to see where the solid ground is.
- 03
Read the evidence
Each card shows the strongest evidence located for that tactic, word for word from the scoreboard.
- 04
Open the study
Follow the links to the underlying paper, vendor study or this site's deeper write-up.
Which GEO tactics actually have evidence?
Of the 14 most-recommended GEO tactics, four have a controlled test behind them: adding quotations, adding sourced statistics, citing external sources and avoiding keyword stuffing, all from one Princeton study. Four more rest on observational data, one (schema) has conflicting results, and five have no study at all. Search below to check any single tactic.
What this lookup does
Most "how to rank in ChatGPT" advice lists every tactic with the same confidence. This GEO tactic evidence lookup lets you check one recommendation at a time against this site's GEO Tactic Evidence Scoreboard, which grades each tactic by the strongest evidence located for it, not by how often it is repeated.
The data is copied from the scoreboard's main table: 14 tactics, their grades, the month they were last regraded, and the best evidence found. Nothing has been added, removed or regraded here. When the scoreboard changes, this tool follows it.
How to read the grades
| Grade | What it means | Tactics |
|---|---|---|
| Fact | Controlled, peer-reviewed or fully disclosed test | 4 |
| Evidence | Disclosed method, but observational | 4 |
| Hypothesis | Widely recommended, no study located | 5 |
| Open | Evidence points in different directions | 1 |
All four Fact-grade rows rest on a single study: the Princeton GEO paper (Aggarwal et al., KDD 2024), a 10,000-query benchmark. It is the one public, peer-reviewed causal test of GEO tactics. Its effect sizes come from 2023-24 testing and predate several current AI search products, so the scoreboard treats the direction as more durable than the percentages.
Worked example: "add schema for AI search"
Say a client report tells you to add JSON-LD schema so ChatGPT and AI Overviews cite you. Type schema in the box. One card comes back, graded Open. Its evidence line says an Ahrefs observational test found no clear lift and that live-fetch tests found schema often ignored. The links take you to the Ahrefs study, the scoreboard's own schema section and the registered schema trial on this site. What you can honestly say afterwards: schema may still be worth it for rich results in classic search, but the claim that it drives AI citations is unsettled.
Now try backlinks. That card is graded Hypothesis: the available data is correlational and reportedly a weaker predictor than brand mentions, with no causal test. Searching matches the tactic name, its evidence line and a few synonyms, so eeat, cwv and json-ld all find the right card.
What it can't tell you
- It doesn't test your page. To check a specific page for the Fact-grade signals (quotes, sourced numbers, citations), use the citation readiness checker.
- It doesn't rank tactics by payoff or give a combined score. Grades are categories on purpose.
- It holds only the scoreboard's 14 rows. If your search returns nothing, the tactic isn't tracked, which says nothing about whether it works.
Grade is not value
An evidence grade tells you how confident to be, not whether something is worth doing. A cheap, reversible tactic with a Hypothesis grade, such as better internal linking, can still be sensible because it may help classic SEO or readers. An expensive tactic with the same grade usually is not. And a Hypothesis grade means untested, not disproven.
Common misreadings
- "Fact means it will work for me." No. It describes evidence quality, not transfer to your topic, competition or engine.
- "Evidence means proven." No. Evidence-grade rows are observational. Correlation is not causation.
- "Open means useless." No. It means studies disagree. Schema markup is the one Open row, and the scoreboard gives it its own section.
- "Not listed means bad." No. The scoreboard tracks the 14 most common tactics only.
Where the links go
Each card links to the sources the scoreboard itself cites for that row: the Princeton paper, the Ahrefs and vendor studies behind the observational rows, and this site's deeper write-ups on content freshness, author signals, schema and llms.txt. Rows the scoreboard marks "no study located" have no study link, because there is none to give.
How grades change
Grades move both ways. A tactic resting on one study drops if a well-run replication fails; a Hypothesis tactic moves to Open if a good test finds nothing. Changes are logged on the scoreboard. Check its "last regraded" column, shown on every card here, before quoting a grade.
Related tools
If the question behind your search is how much AI answers are costing you, the AI traffic loss calculator puts a range on it. For background on the field these grades cover, read what GEO is.
Privacy
The data ships with the page and the search runs in your browser. Nothing you type is sent anywhere.
Last checked and updated: . Everything on this page runs in your browser; nothing you enter is sent anywhere.