Humanity’s Last Choice proposes a GEO evaluation standard, without new experiments

A new paper proposes a common way to evaluate generative engine optimization, or GEO. Humanity’s Last Choice asks researchers to measure whether an edit changes source visibility while preserving the quality of the original user task. It presents an evaluation framework, not evidence that a particular optimization works.

Read the proposal before treating it as a benchmark result

The first arXiv version was submitted on October 16. The paper organizes interventions by the source layer they edit, the action they take and the signal they target. Its proposed protocol tracks permitted edits, source visibility, task performance and edit validity. The authors report no new model experiments, dataset or optimization algorithm in this version.

The framework is a methodological proposal. It does not supply an effect size, a winning page template or proof that adding citations, statistics or keywords improves a publisher’s visibility. The authors also identify limitations in how a taxonomy can represent domain-specific strategies and interacting source layers.

A visibility claim needs a task-quality check

The useful editorial question is whether a more visible source still helps the person who asked the question. A page can attract attention while introducing an unsupported claim. Counting the extra mention alone would miss that failure.

The two cases below are hypothetical review exercises created for this article. They are not experiments from the paper and contain no observed model results.

Write the evaluation rules before editing

1. Define the reader task and the evidence that a satisfactory answer must contain.
2. Save the original source, proposed edit and permitted edit surface. Identify claims that must remain unchanged.
3. Choose and record the engine, model where visible, prompt set, language, account conditions and collection window.
4. Keep visibility, citation support and task-quality observations in separate fields. Preserve failed and incomplete runs.
5. Inspect whether the edit complied with the declared rules before comparing its outcome.
6. Report the sample and uncertainty. Repeat under the same declared conditions before generalizing.

This is a proposed review procedure derived from the evaluation problem, not a completed study. SearchEngineAnswer has not tested an optimization against Humanity’s Last Choice. Any future experiment would need its own source records, scoring rules and observations.

Use the proposal to question reports, not to promise rankings

When a GEO report claims success, ask what changed, what counted as visibility, how the user’s task was scored and which edits were ruled invalid. If those details are absent, the result is difficult to reproduce. A higher mention count alone cannot resolve whether the answer became more useful.

The AI visibility measurement crosswalk separates answer appearances, citations and visits. Use it to keep the visibility measure explicit. For source-level work, the citation-ready passage guide provides a separate checklist for making a verifiable answer easier to inspect.

The publication event is the release of a proposed standard. Until experiments test it, the appropriate claim is that researchers have a framework to examine, challenge and refine.