Envision
Yext Research · Envision 2026

Being the answer

How AI decides which local brand to recommend, and how yours becomes the one it names.
✦
Yext Research · AI search visibility

Four AIs, one question,
four different answers

How the leading AI engines disagree on which local business to recommend, and how your brand becomes the one they all pick.
How it works

AI writes by predicting the next word

At the start, almost anything fits. As the sentence takes shape, the model gets certain. That works when there's one right answer. Ask it who's best nearby, and the guessing shows.
Model's next-word guess ·
One model · the same question · four times

The same model gives the same fact four different ways

what is the capital of lombardy?
ChatGPT · run 1
ChatGPT · run 2
ChatGPT · run 3
ChatGPT · run 4
Where it got the answer

Same forecast, four different weeks, five different sources

Every run agreed: cool mornings, mild afternoons, rain early. They disagreed on which days count as next week, when the rain arrives, and who to cite.
what's the weather in milan next week?
ChatGPT · run 1
ChatGPT · run 2
ChatGPT · run 3
ChatGPT · run 4
When there's no right answer

Ask who's best, and AI gives four different answers

Same question, same canal, four engines. Four different bars on top. Now scale that to two million questions.
where should i get an aperitivo in navigli?
Gemini
    Claude
      Perplexity
        ChatGPT
          So we ran it at scale

          We asked the same local question every way we could

          what are the top 10plumbersnear me?
          0
          distinct keywords
          GeminiChatGPTPerplexityClaude
          4 leading AI engines
          0 markets · monthly
          0
          questions all four engines answeredthe exact set behind every number that follows
          Yext Scout panel · Feb–Aug 2026 · 2.05M local questions asked in total; 491,972 put to all four engines
          The finding
          Same question, same place, same moment. Asked of all four.
          0%
          of the time, all four engines pick the same top business
          0%
          of the time, all four name a completely different one
          And when you ask for a top 10, each engine returns about 7.
          One real query
          No two engines agree on the top pick
          22 different businesses across 23 results. Asked for 10, no engine gave 10, and only one salon makes two lists.
          what are the top 10 health & beauty shops near Castelfiorentino?
          One real query
          All four agree on #1
          16 different businesses across 23 results. All four crown Amplifon, then go their own way.
          what are the top 10 hearing aid stores near Castano Primo, Milan?
          The whole panel, one picture

          How often 3 or 4 engines pick the same business

          Share of questions where at least 3 of the 4 engines name the same top business.
          ✦
          The question behind the question

          How do they define best,
          and who do they consider?

          Best, part one: the reviews

          In traditional search, four stars is the price of entry. Do the engines hold that line?

          Mostly, yes. But the floor is soft, and the models chase the number: 37% of #1 picks are exactly 5.0, often on a handful of reviews rather than the deepest track record.
          4.8★
          median rating of the #1 pick
          52%
          of #1 picks are rated 4.8 or higher
          11%
          still slip in under 4 stars
          The puzzle

          They do reach for rating, just on their own list

          If every engine rewards the same thing, higher ratings, how do they end up agreeing only four percent of the time?
          40%
          of the time, a model makes the highest-rated business on its list the #1
          4%
          of the time, all four land on the same #1
          The resolution

          Same rule. Different rooms.

          Ask four people to point out the tallest person and they will agree, as long as they are in the same room. Every engine reaches for the best-rated name it can see. They disagree because they are looking at four different sets of businesses, not because they judge them differently.
          Claude
          picks its tallest
          Gemini
          picks its tallest
          Perplexity
          picks its tallest
          ChatGPT
          picks its tallest
          ✦
          Getting considered

          So how do you even
          get on the list?

          The consideration gap

          Most businesses never make the list at all

          A typical market holds about 37 local businesses. Across all four engines combined, only about 7 of them ever get named. Before you can win, you have to be considered, and four in five never are.
          7 ever named by an engine
          30 never appear in any answer
          Where each engine looks

          Each engine reads its own list of sources

          Ask the same question and every engine pulls from a different mix of sources. To be considered, you have to show up in the mix it reads.
          First-party sites
          Local websites
          Directories & aggregators
          Industry & government
          Review platforms
          Social, news, forums
          Gemini
          first-party heavy
          42%
          12%
          29%
          13%
          Claude
          reads reviews
          33%
          26%
          11%
          21%
          Perplexity
          most balanced
          32%
          33%
          17%
          ChatGPT
          directory heavy
          27%
          41%
          20%
          Two different games

          Getting listed and getting picked are two different games

          Inclusion is about structure: can the engine read you at all? Ranking is about reviews. Clear the first, or the second never matters.
          1
          Get on the list
          Machine-readable structure, so the engine can find and understand you.
          Schema markupLocation pagesClaimed profileBusiness descriptionComplete attributes
          →
          2
          Win the list
          Reviews and engagement, once you are already in the running.
          Star ratingReview volumeOwner responsesRecent activity
          The inclusion levers

          What actually gets you named

          Lift = how much a signal raises your odds of being named, at the same review level. Above 1.0x helps; below 1.0x hurts.
          Location pages
          2.9x
          Cover photo
          2.8x
          Business description
          2.3x
          Claimed profile
          2.3x
          Schema markup
          2.1x
          Recent posts
          1.8x
          10+ attributes
          1.5x
          Biggest lever, by industry
          Retail & foodLocation pages + schemaup to 3.9x
          Financial & healthcareClaim & complete the profile~3x
          Local servicesKeep posting2.8x
          What consensus looks like

          When all four agree, they're reading the same authoritative sources

          91%
          of unanimous #1s have their own website among the cited sources
          And the category's canonical directory
          The one obvious anchor for a vertical is cited far more often when the engines converge than when they scatter.
          • Waze 2.6×
          • Chamber of Commerce 1.8×
          US panel. Own-site: share of unanimous #1s. Directories: cited this many times more often in unanimous vs split answers. Correlational.
          What it looks like to be #1

          The winners barely out-rate the field. They out-work it.

          A consensus #1 is only a hair higher-rated than a typical business. The real gap is everything else: review depth, freshness, responses, and machine-readable structure.
          Metric
          Unanimous
          Majority
          Typical
          vs typical
          Median reviews
          568
          531
          19
          30×
          New reviews / month
          12.2
          10.4
          1.9
          6.4×
          Google posts, past year
          13.1
          15.5
          6.0
          2.2×
          Photos
          187
          258
          88
          2.1×
          Reviews responded to
          62%
          55%
          33%
          1.9×
          Local-business schema
          56%
          48%
          30%
          1.8×
          Profile completeness
          95%
          93%
          80%
          1.2×
          Avg star rating
          4.2
          4.2
          3.7
          1.2×
          Averages over matched Google Business Profiles. Unanimous n=15,472; typical field n=6.6M.
          Yext Research
          01 / 25
          Yext Research
          1 / 30