'Best X' Questions Make the List Longer, Not Your Share Bigger

What We Compared

We split the unbranded questions tracked in Ranqo into two groups. A best-of question contains "best", "top", "leading", "greatest", "number one" or "#1", or the word for "best" in one of the other languages Ranqo tracks, the rule our prompt generator uses to recognise one. Everything else is an open question: how to solve a problem, what to look for, which approach fits a situation. About 18% of the unbranded questions tracked today are best-of questions.

Answer totals are presented at 10 runs per question per engine, the convention our CRM study uses; every rate, average and share is measured and unscaled.

What this sample is and is not

Two conditions decide which answers count: the engine's current model wrote the answer, and the answer came back with at least one source. Of the engines on Ranqo's entry plan, ChatGPT and Perplexity met both for the whole window.

The companies are the ones tracked in Ranqo, which lean toward software and business services, with an India tilt. It is a small set of companies, so read the within-company results as a direction, not a size. Vendor counts and share of voice use only answers from 6 September 2026 onward, when we raised the number of competitor names our parser records per answer; before that, both were capped and the comparison would be measuring our ceiling. Appearance rates are not capped, so we give them twice: on those same answers, and over the full window, which is large enough to compare each company with itself.

Share of voice here counts every vendor an answer names. The Share of Voice card in Ranqo leaves out names that appear once with no website, so its values run a little higher; counted that way, the result points the same direction. Perplexity's figures span our move to its new API on 25 September, and we kept both sides.

Best-of Questions Get Longer Lists

This part holds up every way we cut it. Both engines name far more vendors when the question asks for the best, and compared company by company the count rose from 4.7 to 7.8, higher for every company.

Almost none of them are dead slots

An answer that names no vendor can never show a brand winning, and a question that keeps producing them is a wasted tracking slot, as our prompt-choice study found. About one open-question answer in 6 (15.7%) named no vendor; about one best-of answer in 15 (6.7%) did. On that measure alone, best-of questions earn their place.

Appearing More Is Not Winning More

The appearance rate rises with the list. On the answers from 6 September 2026, tracked brands appeared in 37.9% of best-of answers against 30.7% of open ones. Over the full window the gap is wider, 41.7% against 29.9%: 45.4% against 30.8% on ChatGPT, and 38.3% against 28.9% on Perplexity. Two things stop that from meaning what it looks like.

Within a company, the rise is smaller

Over the full window, compare each company only with itself and the gap shrinks to 5.6 points (24.0% against 18.4%), against 11.8 pooled, so part of the pooled gap comes from which companies happen to track more best-of questions. The median difference is 0 points, because close to half the companies were never named in either kind of answer. Where a company was named at all, its best-of rate was usually the higher one.

Your share of the names does not grow

Share of voice puts the list length back in the denominator. The brand's share of the vendor names was lower on both engines for best-of answers (5.2% against 6.0% on ChatGPT, 5.4% against 8.0% on Perplexity), and flat within the same companies (3.4% against 3.5%). Leaving out one-off names that come with no website, as the dashboard does, the shares were 6.4% against 7.6%, still lower on both engines. Getting onto a longer list is easier; holding a larger share of it is not.

What to Do With This

Keep best-of questions in the set

Buyers ask them, they almost always produce a list, and the pages AI engines cite most are themselves ranked lists, as our listicle post covers. Dropping them to make a score look steadier would track a market that does not exist.

Judge them by share and position

On a best-of list of seven or eight names, first and seventh are different outcomes that the appearance rate scores the same. Named brands sit in the top five 88% of the time, so a best-of answer that puts you sixth or lower is the exception worth watching.

Compare like with like

Adding best-of questions to a set changes its appearance rate without anything changing in the market, and even surface changes to a prompt move what a model does: one study measured swings of up to 76 points in one open model's accuracy on few-shot tests from formatting changes alone, such as separators and capitalisation, and about 10 points on average across models and tasks. Compare periods only on the same questions, and never compare two brands' visibility when one tracks far more best-of questions than the other. The three decisions behind a share-of-voice number apply here too.

How Ranqo Handles It

A best-of question in each theme, capped below half

When Ranqo writes a brand's first set of questions, it tries to give each theme one best-of question, so every theme has a question likely to return a vendor list, and it aims to keep them under half of the set. That nudges a new brand's visibility up a little by composition alone, which is why we read them by share of voice. The Prompts page shows every question with its category, and its Tracking tab shows, question by question, how often each engine named you. The prompts tutorial walks through it in a few minutes.

Share of voice beside visibility

The Visibility page reports Share of Voice next to visibility, leaving out competitors named only once with no website, so a long tail of one-off names does not shrink every real brand's share. When you act on a gap in the Action Center, the lift it reports shows as a number only once it clears the run-to-run noise we measured; a smaller move reads as no detectable change.