How the CLEO Presence Index works

Methodology, version 1.0

In brief

Buyers now meet a business on five surfaces before they reach its website: Maps, Search, AI Search, Social and Reviews. AI answers compress those surfaces into a short list of names, and a business that is not named is rarely considered. The CLEO Presence Index measures how present a business is across all five surfaces and the website beneath them, on one scale from 0 to 100, from the outside, and in the same way for every business.

Presence is the extent to which a business is found, trusted and recommended across the five surfaces buyers use, resting on a website that machines can reach and read.

Four decisions define version 1.0.

1. It separates what a business builds from what it earns. Presence Readiness, 30 per cent of the index, asks whether search engines, AI systems and buyers can reach and read the business's website and profiles. Presence Recognition, 70 per cent, asks whether search engines, AI answers, reviewers and independent sources find, trust and recommend it. Outcomes carry more weight because outcomes are what buyers see.
2. It counts only what can be attributed. Press releases, wire syndication and paid placements earn nothing on their own. The independent coverage and AI answers they lead to are counted once, weighted by the credibility of the source.
3. It measures every business the same way. Every input is observable from outside and collected from shared question sets for each category and market.
4. It reports its own certainty. Every score shows its coverage and a 95 per cent confidence range. A grade appears only when at least 80 per cent of the method is measured, and movement is reported only when it exceeds measurement noise.

CLEO stewards the index and publishes the full method: signals, weights, scoring curves and rules. Version 1.0 opens a public comment window for version 1.1, with four open questions on which we want the field's best thinking (section 10).

Five surfaces

Buyers meet a business on five surfaces, and AI answers compress them into a few names

Discovery no longer happens in one place. A buyer may meet a business on a map, in a search result, in an AI answer, in a community thread or on a review site, and form a view before visiting its website. CLEO groups these into five surfaces: Maps, Search, AI Search, Social and Reviews. The website sits beneath them as the foundation that machines read; it is not a sixth surface.

AI answers raise the stakes because they compress. Where a results page offers many links, an AI answer usually offers a short list of names, so being named, or not, becomes decisive. Use is growing fast: in BrightLocal's 2026 consumer survey, the share of consumers using AI tools for local business recommendations rose from 6 per cent to 45 per cent in a year [1].

AI answers are also unstable. In research by SparkToro and Gumshoe, the same question put to AI tools repeatedly almost never produced the same list of brands: fewer than one run in a hundred matched. Yet how often a given brand appeared across many runs was stable enough to measure [2]. That finding shapes the whole method: sample widely, report rates rather than single answers, and state the uncertainty.

Existing measures each see one surface. AI visibility tools track answers, search tools track rankings and review platforms track ratings. None tells a business how present it is overall, or where the gap is. The CLEO Presence Index does both: one number, and a breakdown by surface and signal that shows where to act.

Two layers

Presence has two layers: the readiness a business builds and the recognition it earns

The index groups its ten signals into two layers, each answering a question a team can act on.

Presence Readiness (30 per cent) asks whether search engines, AI systems and buyers can reach and read the business. It covers crawl and render access, site content and technical SEO, structured data, social profiles and local listings. These are largely within the business's control, and they change within days of a fix.

Presence Recognition (70 per cent) asks whether search engines, AI answers, reviewers and independent sources find, trust and recommend the business. It covers AI recommendation, AI citation, organic search visibility, reviews and ratings, and independent mentions. These are earned, and they move over weeks and months.

Recognition carries more weight because it is what buyers experience. A technically perfect website that nobody recommends should not score well. Readiness still carries real weight because it is where improvement starts: a site that machines cannot read cannot be ranked, cited or recommended, and Readiness shows the first gains.

Results sit outside the index. Traffic, leads, pipeline and paid reach matter most to a business, but they depend on data only the business can grant, and paid reach can be bought. CLEO reports them alongside the index, from connected data, and never inside it. That keeps the index comparable across businesses and keeps paid media from inflating a measure of organic presence.

Why not owned and earned? Earlier versions split the index into owned and earned presence. That split describes who controls a channel. It does not tell a team what to do next, and several signals sit awkwardly on either side of it. Presence Readiness and Presence Recognition describe cause and effect: build what machines and buyers need, then earn what they give. Readiness is largely what a business owns and Recognition largely what it earns, so the familiar distinction survives, but the labels now say what to do next. Version 1.0 retires the owned and earned labels.

Ten signals

Ten signals, weighted by how directly each shapes what buyers see

Each signal is scored from 0 to 100 and weighted. Weights sum to 100, so a signal's weight is the most it can contribute to the index. Exhibit 3 shows the weights and Exhibit 4 explains each signal in plain language; full definitions are in Appendix A. The construction follows the approach set out in the OECD and European Commission Joint Research Centre handbook on composite indicators: a defined concept, selected indicators, normalisation, weighting, aggregation and robustness testing [9].

Weights in version 1.0 are published judgements informed by the evidence below. We state where the evidence is strong and where it is thin. Before each release, CLEO tests how scores and rankings change when each weight moves 25 per cent either way, and publishes the result. A later major version will calibrate the weights against observed outcomes.

• AI Recommendation carries the most weight (24) because being named as an answer is the outcome buyers see in AI search, and appearance rates across many runs can be measured even though individual answers vary [2].
• Organic Search Visibility (14) remains a distinct surface. Google's guidance is that optimising for its generative AI features is still SEO [3].
• Reviews and Ratings (12) are a primary trust signal for buyers and a source AI answers draw on, particularly for local businesses.
• Independent Mentions (10): in an Ahrefs study of 75,000 brands, branded web mentions correlated more strongly with visibility in Google's AI Overviews (0.664) than backlinks did (0.218) [4].
• Crawl and Render Access (9): AI crawlers other than Google's do not execute JavaScript, so content that appears only after scripts run is invisible to them [5].
• Structured Data (2) is kept as hygiene: in a controlled Ahrefs study of 1,885 pages, adding schema produced no meaningful uplift in AI citations [6].

Attribution

Independent coverage counts; press releases do not

The most common question about any presence measure is whether publicity counts. Version 1.0 answers it directly. A press release is something a business puts out; presence is what others do with it. The index counts the second, not the first.

The evidence points the same way. Muck Rack's May 2026 study of more than 25 million links cited by ChatGPT, Claude and Gemini across 17 industries found that earned media accounted for 84 per cent of citations and journalism alone for 27 per cent, while paid and advertorial content accounted for 0.3 per cent [7]. Earned media's share stayed between 82 and 89 per cent across the study's three editions since July 2025. Press releases themselves were a small and unstable share: about 1.2 per cent of citations in July 2025, about 6 per cent in December 2025, and back near the July level by May 2026, mostly because ChatGPT cited them less often [7]. An academic study from the University of Toronto likewise found a systematic bias in AI search towards earned media over brand-owned content [10].

The measurement profession reached the same conclusion from a different direction. The Barcelona Principles 4.0, published by AMEC in 2025, call for communication to be measured by outcomes and impact rather than outputs, and reject advertising value equivalents [8].

What counts. Independent Mentions counts coverage and discussion from sources the business neither owns nor pays for. Each story counts once, at the credibility tier of the best source that carried it, and its weight halves every 180 days, in line with the finding that most journalism cited by AI is less than a year old [7]. Four kinds of content are excluded before anything is counted: the business's own channels; wire releases and syndicated copies of them; paid, sponsored or advertorial placements; and negative coverage.

The effect is practical. A business improves its score by earning coverage, not by distributing releases, and the Citation Loop in CLEO shows which articles carry it into AI answers.

Measurement

Every business is measured the same way, from the outside

An index is useful only if two scores mean the same thing. Version 1.0 therefore uses only inputs that can be observed without the business's cooperation: public search results, AI answers, review platforms, public web pages and local listings. Data a business chooses to connect, such as analytics or Search Console, powers CLEO's diagnostics and results reporting. It never enters the index.

A standard engine set. The index uses a fixed set of AI engines for every business, weighted by how much each is used, because an answer seen by more buyers matters more; the same logic weights search keywords by volume. The weights follow a published rule rather than judgement (Exhibit 7). Google AI Overviews and AI Mode are held at 25 per cent, because they sit inside Google Search, which trackers of standalone assistants do not measure. The remaining 75 per cent is split across the assistants by their average share of use on two independent trackers, one measuring web visits and one measuring app use [11, 12], with a floor of 5 per cent, and recalculated at every release. Weights follow use rather than referral clicks: presence is about being seen in answers, and clicks are results.

Engines are queried through their providers' official interfaces with web search enabled, or through licensed search results data for Google, without logins or personalisation. Scores therefore reflect what a new buyer is likely to see. CLEO's product tracks more engines for diagnostics; the index uses the standard set so that scores stay comparable.

Lawful data only. The index uses official interfaces, licensed data and public pages CLEO is permitted to fetch. It does not scrape consumer AI apps or platforms whose terms prohibit it. Where lawful, comparable data does not yet exist, as with social engagement and review recency, the index does not measure it yet, and says so.

Certainty

Every score reports how much it rests on and how certain it is

From measurement to score. Each signal starts as a raw measure, such as the share of answers that recommend the business. A published curve converts it to a score from 0 to 100 using two reference points: the level that scores 50 and the level that scores 90. Because the curves stay fixed for the version, a score moves when the business moves, not when its competitors do.

From signals to index. The index is the weighted average of the signals that are live. Coverage is the share of applicable weight that is live. A signal that cannot apply to a business, such as local listings for an online-only business, is excluded from both. Measuring nothing is not the same as measuring zero: a business with no reviews scores zero on Reviews, while a signal CLEO cannot yet measure lowers coverage instead.

Certainty. AI signals are proportions measured from hundreds of answers, so each carries a 95 per cent Wilson interval, the same statistical standard CLEO uses for its GEO Score. With 420 answers from an engine and a true rate of 20 per cent, the interval is about 3.8 points either side. The index is shown with its 95 per cent range.

Movement. The trend line is smoothed to damp day-to-day noise. A change is reported only when it exceeds the minimum detectable change: the smallest difference that measurement noise cannot explain. Comparisons are like for like, using only signals and engines measured at both ends of the period, so a signal or engine coming online never appears as movement. Method changes are marked on the chart.

Business A's surface view shows where to act: Website 83.6, Reviews 72.0, Maps 70.0, Search 60.0, Social 48.1 and AI Search 45.8. Its foundations are strong; its gap is recognition in AI answers.

Business C shows why coverage matters. Its measured signals average 40.2, but a third of the method is not yet measured, so the index is shown as Provisional and no grade is given. When its AI signals go live, the index is recomputed on the full method; because the trend compares like with like, that change in measurement is not reported as a change in presence.

Grades

Grades appear only when the evidence supports them

The bands are unchanged from earlier versions. Two rules make them trustworthy.

• Coverage gating. A grade is shown only when coverage is at least 80 per cent (Rated). Between 50 and 79 per cent, the index is shown as Provisional, without a grade. Below 50 per cent, no index is shown (Unrated) and the live signals are shown on their own.
• Stability. A grade changes only after the smoothed index has stayed at least 1.5 points beyond a band boundary for seven consecutive days, so a business does not flicker between grades on noise.

Scope

What the index tells you, and what it does not

Governance

An open, versioned method keeps the index honest

A measure that sets a category's standard has to be open to challenge. CLEO stewards the index and commits to the rules in Exhibit 11.

Take part

Help shape how presence is measured

Version 1.0 is a start, not a settlement. We invite agencies, search and AI practitioners, researchers, publishers and businesses to test it, challenge it and propose improvements to any part of the method: signals, weights, sources, curves, panels, engines and rules.

What makes a strong proposal. A clear mechanism for why the change would make the index a better measure of presence; evidence from at least two independent datasets or one controlled experiment; and an estimate of how the change would move existing scores. Proposals that would make the index easier to game, or that depend on data that cannot be collected lawfully and comparably, will not be adopted.

The first comment window runs for 30 days from publication.

Appendix

Appendix A. Signal definitions

Scoring curves, minimum-evidence thresholds and the register of platforms for each category are published in the technical specification.

Appendix

Appendix B. Glossary

Appendix

Appendix C. Proposal template

Sources

References

1. [1]BrightLocal, Local Consumer Review Survey 2026.
2. [2]SparkToro and Gumshoe, research on the consistency of AI brand recommendations, sparktoro.com.
3. [3]Google Search Central, guidance on optimising for generative AI features in Google Search, 2026.
4. [4]Ahrefs, analysis of brand mentions and visibility in AI Overviews across 75,000 brands, ahrefs.com.
5. [5]Vercel, with MERJ, The rise of the AI crawler, December 2024.
6. [6]Ahrefs, controlled study of schema markup and AI citations across 1,885 pages, ahrefs.com.
7. [7]Muck Rack, What Is AI Reading?, May 2026 edition, with the July 2025 and December 2025 editions.
8. [8]AMEC, Barcelona Principles 4.0, 2025.
9. [9]OECD and European Commission Joint Research Centre, Handbook on Constructing Composite Indicators: Methodology and User Guide, 2008.
10. [10]Chen, Wang, Chen and Koudas, Generative Engine Optimization: How to Dominate AI Search, University of Toronto, arXiv, 2025.
11. [11]Similarweb, worldwide web-visit share of generative AI assistants, 2026.
12. [12]Sensor Tower, State of AI 2026.
13. [13]Goodie, 2026 AI Search Traffic Report, May 2026.
14. [14]Advanced Web Ranking, Google Organic CTR Study, July 2026 snapshot.

FAQ

Frequently asked questions

Keep reading

Related terms