GEO Investment Study

The decision

If AI visibility work is focused on a single engine today, move it to cross-engine coverage first. A citation earned across all four engines carries roughly 11.6 times the weight of a single-engine citation: the largest, most reliable effect in this dataset, and even more pronounced in independently published research.

Scope and dataset

This edition includes only findings that meet two tests: they hold consistently within our own data, and they are not contradicted by currently published external research. Every number comes from live client accounts only: 1,453 citation records across 18 accounts, tracking the sources cited during question scans across ChatGPT, Claude, Gemini and Perplexity. Internal QA and test accounts are excluded entirely, because one such account was found to distort a headline figure in the prior version. Three findings are stable and corroborated by external research; the fourth is preliminary.

Finding 1: Most citation volume goes to sources unrelated to any one brand

• Unrelated (competitor or category) sources: 6,285 total weight across 1,391 records, 95.1% of all citation volume.
• Earned coverage: 311 weight across 59 records (4.7%). Owned content: 10 weight across 3 records (0.2%).
• Within the unrelated pool, other companies' own sites account for 62% of the weight. Of all vendor-site citations, 99.3% belong to other companies and 0.2% to the brand's own site.
• No single rival holds more than roughly a fifth of the field. On average, the biggest competitor captures less than 8% of the citations a brand does not win, and the top three combined average under 18%.

Engines answer category questions with the sites of companies that sell in that category, and the field is fragmented enough to be genuinely open. The lever is presence on the third-party pages AI already pulls from, not only polishing a brand's own site.

Finding 2: A linked, page-level citation carries about 2.5 times the weight of a brand mention

• Citations that link to a specific page: mean weight 5.21 across 1,145 records.
• No specific page cited: mean weight 2.06 across 308 records.

This is about how a page is written, not how many pages exist. Pages that open with a direct, quotable answer outperform pages built around general brand narrative. Half of cited pages in this dataset are homepages (Finding 4), so one well-structured page can answer many questions.

Finding 3: Cross-engine presence beats depth on one engine

Mean citation weight rises sharply with the number of engines a citation appears on. By engine, the spread is modest by comparison: Gemini 7.69, Claude 7.34, ChatGPT 6.62, Perplexity 6.18.

Report AI visibility per engine, not as one blended score, and widen coverage to the engines currently ignored before going deeper on the one already working. Published 2026 studies find even lower overlap between engines than this, so the finding is, if anything, conservative.

Finding 4 (preliminary): Half of cited pages are homepages; FAQ pages are 4 of 4,908

This does not mean FAQ content is worthless, or that Finding 2 is wrong: Finding 2 measures whether a citation links to a specific page, a different question from what format that page takes. Roughly a third of page paths could not be classified, and the full fetch-and-classify pass has not been done. Treat this as a strong early signal: do not over-invest in FAQ formatting on faith, and make homepages and core pages exceptionally clear.

What this means, in priority order

1. Go cross-engine before anything else: secure presence on ChatGPT, Claude, Gemini and Perplexity before adding net-new content.
2. Build pages that lead with the answer: rewrite briefs so pages open with a direct, quotable answer instead of brand narrative.
3. Assume most of the category's citation volume is not yours to start, and budget for expanding owned and earned presence into an open field.

The study measured no cost or ROI, so it proposes priorities, not a spend ratio.

External research each finding was checked against