Three Streams GEO Methodology
Content Scorecard
Score one page on eight quality checks for a pass or fail, then, if you like, six writing techniques for a publish verdict. It is a quality standard, not a citation predictor: it finds quality gaps on a page and does not forecast citations.
What this scorecard is
A quality standard, not a citation predictor. Use the checks to find and fix quality gaps on a page.
Before scoring, two checks: the page is indexed by the search crawlers, and every citable fact is in the served HTML. Then two parts. Gate 1 scores eight quality checks, pass or fail; Gate 2 is optional and scores six writing techniques out of 30, which together with Gate 1 gives a publish verdict.
Each check and technique is tagged with the stage of an AI answer it acts on: 1 query processing, 2 retrieval, 3 selection, 4 the answer.
Why a standard, not a prediction
Within a single site, page properties of this kind had no significant effect on citation; retrieval did π Chen & Koudas, arXiv 2610.15033, 10 Oct 2026; preprint; API models, not consumer apps.
What the 32% and the 2.8% mean
In a 2026 study, how often and how high a page was retrieved predicted about 32% of the differences in citation counts between pages of the same website; the pageβs own features predicted about 2.8% π Chen & Koudas, arXiv 2610.15033, Oct 2026; API models, not consumer apps.
The 32% and the 2.8% come from separate analyses and cannot be added. The 2.8% is all the measured page features combined, not any single feature. Content also acts through retrieval: a page is retrieved because it matches the searches, so much of contentβs effect sits inside the 32%. Page quality therefore cannot promise citation, but it is not negligible.
Where the eight checks come from
Each check has a job in getting a page found, selected or used in the answer, and its screen shows the evidence behind it. None is proven to cause citation on live engines: the evidence is correlational or descriptive.
* The checks originate in GEO-16, a page-quality framework π Kumar & Palkhouski, arXiv 2509.10762, Sep 2025; preprint; 70 prompts, B2B SaaS, three engines without ChatGPT. The r values are its correlations with citation across websites: associations, not effects. TSM keeps the checks that have a job in retrieval and current evidence, and scores no others.
How it is scored
Score each check from 0 to 3. A check scored 2 or 3 passes. To pass Gate 1, a page needs at least 70% of the available points and at least 75% of its checks passing: 17 of 24 points and 6 of 8 checks on a standard page.
On an operational page (returns, shipping, warranty, rewards), Sourced claims does not apply, so seven checks are scored: 15 of 21 points and 6 of 7 checks. In Gate 2, three techniques apply to operational pages and the pass mark is 12 of 15. Which items apply is set by the page type, not chosen page by page.
TSM default; calibrate to your data.
What is not scored here
Schema is entity clarity and belongs to the technical checks; it has no proven citation effect. Author credentials are checked in the E-E-A-T review. Disclosures are checked in the compliance review.
What each score means
| Score | Meaning |
|---|---|
| 0 | Not present |
| 1 | Partially implemented |
| 2 | Implemented (the check passes) |
| 3 | Fully implemented |
Score descriptions are a TSM operationalization. A check you skip counts as 0.
How Gate 2 works
Gate 2 scores six writing techniques Gate 1 does not cover: Direct Answer Format, Entity-First Writing, Technical Terms & Definitions, Fluency Optimization, Expert Quotation Integration and Topic Match, 5 points each. Topic Match is a TSM addition from 2026 evidence, not a technique from the library. Each technique is split into criteria worth 1 or 2 points; a criterion you leave open counts as 0.
Gate 2 is optional in this scorecard. Without it you get the Gate 1 result only; with it you get a publish verdict.
Minimum publication threshold
Found and readable, Gate 1 pass AND Gate 2 at 77% of the available points or more: 23 of 30 on a standard page. Gate 1 is pass or fail, is not scored in points and cannot be offset by Gate 2. The two gates are never added together.
| Gate 2 score | Result | Action |
|---|---|---|
| 27β30 (90% or more) | Excellent | Publish as-is or minor tweaks |
| 23β26 (77% or more) | Good | Address priority issues before publishing |
| 16β22 | Needs improvement | Significant rewrite needed |
| <16 | Poor | Complete rewrite required |
On an operational page three techniques apply (Direct Answer Format, Entity-First Writing and Topic Match), out of 15: 14β15 excellent, 12β13 good, 8β11 needs improvement, under 8 poor.
Gate 2 points, groups and the 77% minimum: TSM default; calibrate to your data.