From Rankings to Recommendations: Measure Presence, Not Position

Our opinion, plainly: presence is the unit, not position

This is ContextOwl's point of view, built from running one-time AI visibility reports — we have a commercial stake here, and we say so plainly rather than burying it.

Our claim: a single "rank" is close to meaningless, because the same prompt rarely returns the same list twice. What repeats is whether a brand gets named at all, across many runs — so the number worth reporting is an appearance rate, not a position borrowed from the SERP era. This isn't yet standard; most dashboards still show a "rank" per prompt because it's familiar. We think that number is often closer to noise than signal — see can you actually rank #1 in ChatGPT? for the question asked directly.

What actually changed, from rankings to recommendations

A results page shows ten links and lets the reader scan and choose. Position is real and countable there, because the page doesn't change between two people searching the same term seconds apart.

An AI answer is assembled per request instead — the model, and whatever sources it pulls that moment, compose a short narrated recommendation rather than read off a fixed list. That's where the instability comes from: candidate brands may be the same, but ordering, count, and even which brands make the cut can shift run to run, with identical wording. Google describes its own AI features this way — retrieval and synthesis from its index, not a static ranked set; see our Google AI search optimization guide. What you're measured against is a recommendation being generated, not a page being served.

Why presence, not position, is the honest unit

The strongest evidence: SparkToro, with Gumshoe.ai (a commercial AI-tracking tool — a disclosed conflict), ran roughly 2,961 brand-recommendation prompts across ChatGPT, Claude and Google's AI, 60–100 times each, in categories like chef's knives and headphones. Via Search Engine Journal and Search Engine Land; SparkToro's own post wasn't accessible, so treat the figures as secondary-sourced.

Findings: an identical list twice was below a 1% chance; identical order below 0.1%. But frequency held up — top brands still appeared in roughly 55% to 77% of responses for a given intent, regardless of phrasing. Order is close to random; presence is comparatively stable. "We rank #2" may not survive the next run. "We appeared in 68% of 50 runs" likely will. See how many runs that actually takes for the full breakdown of this same study.

Here's what tracking that over time would look like:

Illustrative example only, for a fictional brand ("Nimbus Cloud Storage") with invented numbers — not real scan data. It shows the shape of a presence-over-time view: the appearance rate is the trend line to watch; the position range is context, not the headline.

Mentions as the KPI, as clicks decline

Appearance rate matters more for a second reason: the click used to attribute value is fading too. Pew Research Center, studying ~900 US adults' browsing in March 2025 (published 22 July 2025), found that when a Google AI summary appeared, people clicked a traditional result link in 8% of visits, versus 15% without one. Clicks inside the summary itself happened in just 1% of visits, and 26% of summary visits ended with no further click, versus 16% without.

That's Google AI Overviews specifically, measuring clicks, not what answers said about any brand. But it's a real sign the click is weaker than it used to be. If fewer people click, a brand mention — named, described, ideally recommended — becomes the thing you can verify happened, closer to how AI search visibility is defined: by what the answer says, not what the reader does.

What this implies for how teams report results

• Report a rate over N runs, not a rank from one. "Named in 32 of 50 runs" survives scrutiny; "we're #1" doesn't.
• Treat position as secondary — log it, don't lead with it.
• Compare share of voice across a prompt set; see our walkthrough on measuring AI visibility.

The strongest case against this view — stated fairly

The best objection isn't "position doesn't matter" — it's that one appearance-rate number can erase a real signal. A brand named in 70% of runs but always listed last may lose business to a rival named in only 55% of runs but almost always first and recommended outright. Primacy plausibly shapes choice much as the top of a results page does. A headline number that ignores where and how you were named just trades one oversimplified metric for another.

We think that objection holds. Position isn't worthless — only noisy enough, run to run, that it shouldn't be the headline, and should be reported alongside sentiment and whether the brand was recommended, not just named.

What would actually change our mind

Two things. First, a study showing position is more stable than the SparkToro/Gumshoe.ai figures suggest, across more categories and engines. Second, evidence that position, not presence, predicts a downstream outcome — a click, a signup, a sale — with presence held constant. If being named first reliably converts better than being named often, we'd reweight our reports.

FAQ

Is "presence over position" a proven fact? No — it's ContextOwl's opinion, and we sell AI visibility reports, so we have a stake in this framing. We've tried to state the counterargument fairly rather than dismiss it.

What's the difference between visibility and ranking? Ranking assumes a fixed, orderable list. Visibility asks how often an engine names, describes, recommends or cites you across many runs — see our plain-English definition.

If order is nearly random, why log position at all? Because always-first and always-last are different competitive positions, even if neither is the headline number. Log it; just don't lead with it.

What to do next

1. Run one buyer-intent prompt 10–20 times in fresh chats on at least one engine. Log named/not-named before touching position.
2. Calculate the appearance rate (named ÷ total runs); note position only as context. See our measurement walkthrough and audit checklist.
3. Comparing tools? Check whether each reports a rate across runs or a single-run rank — see how to choose an AI visibility tool.

ContextOwl sells one-time AI visibility reports, not a monitoring subscription, and this framing supports that product — see a multi-prompt, multi-run read in the AI Visibility Report sample.