Sameness Index · 4 sites compared

Your page scores 38 out of 100 for sameness against the 3 competitors you named — low enough that buyers can probably tell you apart.

https://autoscience.ai/autoresearch

Share score cardDownload as markdown
01

Your verdict

38
Sameness · vs 3 named
How is this calculated?

Your page is mostly your own.

Most of what you say, Weco and the rest of your set do not. You're less same than 340 of the 366 SaaS sites scored (SaaS avg 56).

No category benchmark matched this set.
Sameness Index on a 0 to 100 scale, from distinctive at 0 to interchangeable at 100. This page scores 38. Weco scores 51. Pavo scores 41. Coral Autoresearch scores 56. The SaaS avg is 56.

Each named site is scored the same way, against the other 3 in this set, so its tick means the same as your marker. The dashed line is the frozen benchmark average.

  1. Less same than average

    The average SaaS site scores 56; you scored 38.

    You are 18 points less same than the average SaaS site.

  2. Closest overlap: Weco

    Of the 3 sites you named, Weco echoes the most of what your page says: 54% of it, weighted by placement. Scored the same way you were, against the rest of the set, Weco's own Sameness Index is 51. Those are different measures: the first is overlap with your page, the second is how same Weco's whole page is.

    Weco is the competitor you sound most like.

  3. Room to own more

    28% of your claim space is ownable: unique, relevant, and hard to copy. 6 of those claims sit in body copy, where few readers reach them.

    28% is ownable, and 6 buried opportunities could help you stand out more.

Do this first

Three changes worth testing first.

Chosen by rule from the comparison with Weco, Pavo and Coral Autoresearch: the shared claim taking your most prominent space, then the claims only you make that sit too low on the page to be read. Each one links to its claim card.

  1. “It identifies promising research, executes experiments within your stack, and ships verified pull requests that beat your baseline”

    Weco, Pavo and Coral Autoresearch all say it too. Buyers may still need it, but shared ground cannot carry your hero — move it lower and give that space to something only you can say.

    Table stakes
    Most of the set says this too.
  2. “Your model and data remain yours; with on-prem deployment, nothing leaves your network”

    Nobody in the set says this. It sits in body copy, where few readers reach it — worth testing higher up the page; only buyers can tell you whether it lands.

    Surface
    Yours alone. Test it higher up.
  3. “Does not replace researchers; runs experiments your researchers don't have bandwidth for and reports results, team reviews and steers”

    A claim that is yours alone, filed in body copy. Try it where it will be read before the shared claims are, and let buyers tell you if it moves them.

    Surface
    Yours alone. Test it higher up.

Try these changes, then test them with real buyers.

This measures overlap. Whether buyers notice is a different question, and only they can answer it.

Test it with real buyers
02

What you can own

Claims only you make, that buyers weigh, and that competitors can’t easily copy.

28% of your page’s claim space is yours to keep.

28%Yours to keep31%Unique but weak41%A competitor says it too
Why this is not 100 minus the Sameness Index

The index is a weighted composite across six categories, including page structure and visuals. This bar is measured on your claims alone, weighted by where each one sits on the page. Different denominators, so the two never add to 100 and are not meant to.

Already leading with · 4

  1. Deployable on-prem or in your cloud

    “Deploy in the cloud or on-prem”

  2. Benchmark ranking in public ML competitions

    “96th percentile in public ML competitions”

  3. Ingests published research and applies it to your model

    “Mira incorporates the most relevant, novel research into your model”

  4. Outperforms human ML researchers autonomously

    “Mira autonomously outperforms nearly every human at ML research”

Buried in body copy · 6

  1. Improves inference speed

    “Speed: Faster inference”

  2. Reduces compute cost/resource use

    “Cost: Resource efficiency”

  3. Augments rather than replaces your researchers

    “Does not replace researchers; runs experiments your researchers don't have bandwidth for and reports results, team reviews and steers”

  4. Improves model quality/accuracy

    “Quality: Smarter models”

  5. Customer retains ownership of model and code

    “You own the model and the code; nothing is licensed back”

  6. Your data never leaves your network

    “Your model and data remain yours; with on-prem deployment, nothing leaves your network”

03

Where you blend in

Territory you spend prominent space on that the set also occupies. Not every line is one to delete — the question is whether it has earned the space, or whether something only you can say should be there instead.

  • Commodity · 100%Table stakesHero
    autonomous AI agent runs ML research/experiments
    • “It identifies promising research, executes experiments within your stack, and ships verified pull requests that beat your baseline”
    • “Autoresearch for Frontier Teams”
    • “24/7 experiments in flight”

    You say this 3 different ways.

    Weco, Pavo and Coral Autoresearch all cover this territory. Buyers may need to hear it, but in your hero it spends the first impression on shared ground.

    They say
    • Weco“Production-grade AutoResearch for your Models Agents Prompts Configs”
    • Pavo“Pavo develops and evaluates competing changes across prompts, workflows, heuristics, models, training, and optimization, then carries the strongest candidates…”
    • Coral Autoresearch“AI agents can propose ideas, test them and improve them. That's automated research.”
  • Commodity · 100%Table stakesSection
    beats or exceeds human researcher output
    • “Mira autonomously outperforms nearly every human at ML research”
    • “Evaluate all the ideas that you normally wouldn't have capacity to”

    You say this 2 different ways.

    Common ground with Weco, Pavo and Coral Autoresearch. Say it if buyers need it — lower on the page, where it is not the thing they read first.

    They say
    • Weco“Weco consistently surfaced improvements across my entire ML pipeline that I would never have discovered manually. It doesn't just accelerate development - it…”
    • Pavo“Every production system has more ways to improve than any team can explore manually.”
    • Coral Autoresearch“See how focused agents solve more tasks.”
  • Commodity · 100%Table stakesSection
    benchmark and leaderboard results as proof
    • “96th percentile in public ML competitions”

    Weco, Pavo and Coral Autoresearch make the same claim. It cannot set you apart, so it should not carry the section. Nobody else uses your exact words, but everyone is on the ground. Nobody else makes this exact claim, but everyone occupies the territory — lead with the specific, not the generic.

    They say
    • Weco“Fraud Detection: AUC 0.91 → 0.93 for end-to-end transaction-fraud ML pipelines”
    • Pavo“Learned router + validation heuristic selected for production, benchmarked +14%, guardrails cleared, experiment design and staged rollout plan attached.”
    • Coral Autoresearch“62 fewer cycles for the same calculation”
  • Commodity · 67%Table stakesSection
    optimizes whatever metric you define
    • “You name the metric. Mira optimizes for it.”

    Weco and Pavo make the same claim (67% of the set). It cannot set you apart, so it should not carry the section.

    They say
    • Weco“Optimize your code against a measurable metric, like speed, accuracy, and cost.”
    • Pavo“Finds where the metric is stuck - which surface, which cohort: feed or related, new users or lapsed.”
  • Contested · 33%SharpenHero
    only validated improvements ship
    • “Ships verified pull requests that beat your baseline”

    Pavo is on this territory too (33% of the set). It narrows the field without winning it — make it specific enough that it cannot be said of them.

    They say
    • Pavo“Explores widely, and ships only what's proven.”
04

Claim-by-claim evidence

Every claim on your page (22)
What the columns mean
Claim
The grouped claim, then your exact line beneath it.
Type
What kind of claim it is: category, segment, outcome, capability, quality or proof.
Placement
Where it sits on your page: hero, section or body copy. Hero claims weigh most in the index.
Same claim
Share of the competitors making this exact claim. Drives ownership and ownable share.
Same territory
Share of the competitors with any claim in the same buyer-facing territory. This is what the index is scored on.
Sayability
Whether a competitor could truthfully make the same claim: anyone could, copyable with effort, or hard to copy.
Relevant
Whether buyers decide on this. A unique claim nobody buys on is not ownable.
Ownership
Commodity: 60% or more of the set says it. Contested: 20–59%. Unique: under 20%, owned when it is also hard to copy.

Tap a column to sort by it; tap again to reverse. Sorted by Same claim, highest first.

  • Agent proposes, runs experiments, ships code changes
    “It identifies promising research, executes experiments within your stack, and ships verified pull requests that beat your baseline”
    capabilityHerosame claim 67%same territory 100%Copyable with effort
    Commodity
    Also on Pavo, Coral Autoresearch
  • Positions as autoresearch/automated research category
    “Autoresearch for Frontier Teams”
    categoryHerosame claim 67%same territory 100%Anyone could say it
    Commodity
    Also on Weco, Coral Autoresearch
  • Only submits changes that beat the baseline
    “Ships verified pull requests that beat your baseline”
    proofHerosame claim 33%same territory 33%Copyable with effort
    Contested
    Also on Pavo
  • Explores more ideas than the team has capacity for
    “Evaluate all the ideas that you normally wouldn't have capacity to”
    outcomeSectionsame claim 33%same territory 100%Anyone could say it
    Contested
    Also on Pavo
  • Optimizes any metric you specify
    “You name the metric. Mira optimizes for it.”
    capabilitySectionsame claim 33%same territory 67%Anyone could say it
    Contested
    Also on Weco
  • Setup needs only repo access and a metric
    “Only needs repo access and a description of your evaluation metric to get started”
    qualityBodysame claim 33%same territory 33%Anyone could say it
    Contested
    Also on Weco
  • Deployable on-prem or in your cloud
    “Deploy in the cloud or on-prem”
    capabilityHerosame claim 0%same territory 0%Copyable with effort
    Unique and owned
  • Benchmark ranking in public ML competitions
    “96th percentile in public ML competitions”
    proofSectionsame claim 0%same territory 100%Copyable with effort
    Unique and owned
  • Ingests published research and applies it to your model
    “Mira incorporates the most relevant, novel research into your model”
    capabilitySectionsame claim 0%same territory 0%Copyable with effort
    Unique and owned
  • Outperforms human ML researchers autonomously
    “Mira autonomously outperforms nearly every human at ML research”
    outcomeSectionsame claim 0%same territory 100%Copyable with effort
    Unique and owned
  • Runs experiments continuously, 24/7
    “24/7 experiments in flight”
    capabilitySectionsame claim 0%same territory 100%Anyone could say it
    Unique for now
  • Thousands of new ML papers published weekly
    “~2,000 new ML papers weekly”
    proofSectionsame claim 0%same territory 0%Anyone could say itnot a buying criterion
    Unique for now
  • You never miss relevant new research
    “Never Miss an Important Paper Again”
    outcomeSectionsame claim 0%same territory 0%Anyone could say itnot a buying criterion
    Unique for now
  • Adds new model capabilities/features
    “Capability: New feature support”
    outcomeBodysame claim 0%same territory 33%Anyone could say itnot a buying criterion
    Unique for now
  • Augments rather than replaces your researchers
    “Does not replace researchers; runs experiments your researchers don't have bandwidth for and reports results, team reviews and steers”
    qualityBodysame claim 0%same territory 33%Anyone could say it
    Unique for now
  • Customer retains ownership of model and code
    “You own the model and the code; nothing is licensed back”
    qualityBodysame claim 0%same territory 0%Anyone could say it
    Unique for now
  • Improves inference speed
    “Speed: Faster inference”
    outcomeBodysame claim 0%same territory 67%Anyone could say it
    Unique for now
  • Improves model quality/accuracy
    “Quality: Smarter models”
    outcomeBodysame claim 0%same territory 33%Anyone could say it
    Unique for now
  • Improves sample efficiency of data
    “Data: Sample efficiency”
    outcomeBodysame claim 0%same territory 33%Anyone could say itnot a buying criterion
    Unique for now
  • Produces smaller models
    “Footprint: Smaller models”
    outcomeBodysame claim 0%same territory 67%Anyone could say itnot a buying criterion
    Unique for now
  • Reduces compute cost/resource use
    “Cost: Resource efficiency”
    outcomeBodysame claim 0%same territory 67%Anyone could say it
    Unique for now
  • Your data never leaves your network
    “Your model and data remain yours; with on-prem deployment, nothing leaves your network”
    qualityBodysame claim 0%same territory 0%Copyable with effort
    Unique and owned
05

How this was calculated

Sameness measures how much your claims overlap with the sites compared. It does not measure message quality or whether buyers prefer you.

AI-analyzed: an AI read each page on its own and grouped the claims that say the same thing. No score here was written by a model — every number is computed from those groupings in our own code, with the weights below.

How the score is built
The six category scores, their weights, and what a high score in each one means
CategoryWeightYoursWhat a high score means
Messaging
Category framing, who it is for, and the outcome promised
30%70The most expensive kind of sameness. A buyer cannot tell what job you do that the others do not.
Claims
Attribute and benefit claims — speed, ease, quality, ROI
30%17Every shared claim is a line already read on another tab. Cut the ones nobody owns and spend the space on something they cannot.
Features
Capabilities and functions the page lists
15%53Expected in a mature category, and the least alarming of the six. Feature parity is normal; leading with it is the mistake.
Proof
The kinds of evidence offered: customer logos, numbers, testimonials, case studies, badges
10%21Same kinds of proof as everyone means the proof stops working as proof. It is scored on the kind of evidence, not on which customers are named.
Structure
Section order, navigation, CTA language and placement
10%0The generic SaaS template — hero, logos, three-feature grid, testimonial, CTA. Familiar is not the same as memorable.
Visual
Palette family, imagery style, layout patterns
5%44Weighted lowest on purpose: buyers rarely decide on this. Worth knowing, rarely worth fixing first.

Each site was read on its own first, with no knowledge of the others, so your page gets no benefit of the doubt a competitor’s does not. A category nothing could be measured for drops out and the rest are re-weighted, rather than counted as zero.

What we compared (4 pages read)
Your page
Mira (by Autoscience)
autoscience.ai
Competitor
Weco
weco.ai
Competitor
Pavo
pavoai.com
Competitor
Coral Autoresearch
coral-labs.ai
What it cannot tell you

The index can find where two pages converge. It cannot say whether a buyer would notice, or which of your reasons to buy actually land. A single check also moves several points between runs, so read the band and the ranking, not the last digit.

Your highest-impact changes

  1. 1
    Table stakesYour hero copy says “It identifies promising research, executes experiments within your stack, and ships verified pull requests th…”.

    Weco, Pavo and Coral Autoresearch all say it too. Buyers may still need it, but shared ground cannot carry your hero — move it lower and give that space to something only you can say.

  2. 2
    Table stakesYour section copy says “96th percentile in public ML competitions”.

    Keep the fact, lose the position: weco, Pavo and Coral Autoresearch all say it too, and your section is spending its first impression on the same territory as theirs.

  3. 3
    Table stakesYour section copy says “Evaluate all the ideas that you normally wouldn't have capacity to”.

    This is the set's common ground — Weco, Pavo and Coral Autoresearch all say it too. It will not set you apart wherever it sits, and in the section it costs you the one place a distinctive claim would be read.

  4. 4
    Table stakesYour section copy says “Mira autonomously outperforms nearly every human at ML research”.

    A buyer with three tabs open reads a version of this on every one of them. Say it further down for the readers who need it; the section should carry a claim they will only find here.

  5. 5
    Table stakesYour section copy says “24/7 experiments in flight”.

    True of you and true of them: Weco, Pavo and Coral Autoresearch all say it too. That is why it decides nothing, and why the section is the wrong place to spend it.

  6. 6
    Table stakesYour section copy says “You name the metric. Mira optimizes for it.”.

    In the section: Weco and Pavo say it too (67% of the set). Buyers may still need it, but shared ground cannot carry your section — move it lower and give that space to something only you can say.

  7. 7
    Table stakesYour body copy says “Speed: Faster inference”.

    In the body: True of you and true of them: Pavo and Coral Autoresearch say it too (67% of the set). That is why it decides nothing, and why the body is the wrong place to spend it.

  8. 8
    Table stakesYour body copy says “Footprint: Smaller models”.

    In the body: Keep the fact, lose the position: pavo and Coral Autoresearch say it too (67% of the set), and your body is spending its first impression on the same territory as theirs.

  9. 9
    Surface“Your model and data remain yours; with on-prem deployment, nothing leaves your network” is yours alone, and buyers weigh it.

    Nobody in the set says this. It sits in body copy, where few readers reach it — worth testing higher up the page; only buyers can tell you whether it lands.

  10. 10
    Surface“Does not replace researchers; runs experiments your researchers don't have bandwidth for and reports results,…” is yours alone, and buyers weigh it.

    A claim that is yours alone, filed in body copy. Try it where it will be read before the shared claims are, and let buyers tell you if it moves them.

  11. 11
    Surface“You own the model and the code; nothing is licensed back” is yours alone, and buyers weigh it.

    No competitor page makes this claim. Today it is in body copy; it is a candidate for the space the table-stakes lines are taking.

  12. 12
    Surface“Speed: Faster inference” is yours alone, and buyers weigh it.

    The rest of the set has nothing on this ground. Buried in body copy it does no work; higher up it is the one line a comparing buyer cannot find elsewhere.

  13. 13
    Surface“Quality: Smarter models” is yours alone, and buyers weigh it.

    Filed in body copy: Nobody in the set says this. It sits in body copy, where few readers reach it — worth testing higher up the page; only buyers can tell you whether it lands.

  14. 14
    Surface“Cost: Resource efficiency” is yours alone, and buyers weigh it.

    Filed in body copy: Nobody in the set says this. It sits in body copy, where few readers reach it — worth testing higher up the page; only buyers can tell you whether it lands.

The only way to know if it matters.

This report can tell you where your messaging overlaps. It cannot tell you whether a buyer would care, or which of your reasons to buy actually land. Put the page in front of real B2B buyers in your target market and ask them.

Test it with real buyers
Trusted by
HubSpotRingCentralShopifyCognismPaddleVeeamRipplingMiro