/ The Bottom Line Up Front
A model can know your name and still describe you as the wrong kind of company.
Recognition and correct description are two different measurements. In the 1,000-brand study behind this tool, 510 brands were volunteered unprompted and 194 were filed under the wrong category. Zoho is both: named readily, read well, and ranked as CI/CD rather than CRM.
What each probe measures
The audit reads an open model directly and reports what each probe found:
- Whether the model volunteers your brand unprompted, and which category it files you under
- How it regards you on twelve buying dimensions, read from activation space rather than generated text
- How far your brand name surfaces as the signal travels up the model’s layers
How the ModelRank brand audit works
Name your domain
We read a handful of your pages to place you in the category taxonomy, and you can name up to seven competitors or let the crawl suggest them.
Four probes run live
Recall, category binding, activation-space sentiment and the logit lens. Each one streams back the moment it lands, so the report fills in as it is measured.
The channel phase queues
Third-party surfaces, backlinks and own-property facts for you and every named competitor, measured per brand and placed as an ordinal among the peers you named.
The audit keeps a link
Every run is stored under its own key, so the finished report can be reopened and shared without running the model again.
What the ModelRank audit looks like
A real audit for Stripe, rendered by the same components your own report uses. Your audit measures your domain live and produces a report of exactly this shape.
ModelRank/payments
Stripe
Recitable
deeply imprinted, reproduced on demand
in payments, the rank that matters
The model holds this brand, volunteers it, and files it under the right category.
What to do about this3 actions from this audit, in priority order
- 1.
Name third-party integrations on the features page
Stripe has no integrations page the crawl could reach.
Memory depth → - 2.
Publish reference documentation that answers a task end to end
The model reads Stripe below its named peers on documentation.
Sentiment → - 3.
Say what the product is in the opening sentence of every page you own
The brands at the core of payments state their category in their first sentence; Stripe does not.
Where you sit →
from the recall and imprint probes
See how to action this
the share of your measured surfaces that also state what you do
See how to action this
mean read across the twelve buying dimensions
See how to action this
from the corrected length-matched fact probe
See how to action this
Measured on google/gemma-4-12B on one date; every model holds a different imprint of the same brand. Read the method
How much of Stripe the weights hold
M, the first of the four parts of the score, and the one with the most weight on it. Two probes measure it: whether the model volunteers the name at all when asked about the category, and how much of the brand's own published surface is reproducible from the weights.
The channel composites, each a bundle of measured columns rolled into a single position among the competitors this audit named. These are the parts of the picture a content plan can actually move, and any action the audit planned for a lever sits under its bar.
- Category explanation5th of 6
Built from how often the model has seen the brand name and its category described together.
- Review sites4th of 6
Built from pages on G2 and Trustpilot, whether those pages explain what the product does, and review volume.
- Own properties1st of 6
Built from whether a help centre exists and how many articles it has, how much documentation is crawlable, whether pricing is disclosed, whether there is a free tier and self-serve signup, and how many integrations are listed.
Which of these pages the weights absorbed- features-2.17 on 1 passagesingle
- home-2.52 on 1 passagesingle
- pricing-1.90 on 1 passagesingle
- Documentation-1.68 on 1 passagesingle
The crawl did not capture enough blog and engineering writing or homepage and pricing pages to score those kinds, so they are unscored rather than absent from the weights. The scored surfaces rest on 1, 1, 1 and 1 passages respectively. Those are small samples, and the rows are printed in a fixed order rather than ranked, because differences of this size between one brand's own surfaces are mostly noise. 4 rows rest on a single passage and is flagged as such. Nothing here establishes that one kind of page imprints better than another, for Stripe or for anyone.
Stripe has no integrations page the crawl could reach.Close thisName third-party integrations on the features page
The features page opens with content regarding sessions and global payments, mentioning third-party integrations generally without listing them. Provide a dedicated section or link that names the specific third-party tools and applications compatible with the platform.
https://stripe.com/features
How to read these barsPlaced against category peers, not a fixed population; a missing composite is missing data
Each bar is where Stripe places among the measured payments brands carrying the same composite, not among a background population. A live audit measures a customer and the competitors they name, so a peer placing is the comparison that survives into the product. A composite neither Stripe nor enough of its peers carry is left without a bar rather than drawn as a zero, because a zero-length bar reads as a bad score rather than as a missing measurement.
- Named unprompted30/40
Stripe is volunteered when the model is asked about payments.
- Your own pages in the weights25/25
The brand's own pages score -2.07 on the surface imprint probe.
How M is builtThe two probes rescaled onto 0 to 100, giving 84.6
The two probes are added and rescaled onto 0 to 100, which gives M = 84.6. Among the 6 measured payments brands that puts Stripe at the 83th percentile on memory depth specifically, which is a different question from where the composite puts it. This is the part that moves slowest: weights only change when a model is retrained.
What the model associates with you
Each node is a named concept axis, fitted from hand-written contrastive prompt pairs before any brand was measured, so the axis means what it says rather than whatever an unsupervised feature happened to latch onto. Node area is how firmly the model holds that axis for Stripe; node state is what to do about it. Their agreement with the payments axis profile is C in the score: 50.0 of 100, read here from the share of Stripe's measured surfaces that also state what it does, which is the live equivalent of the cohort's concept alignment rather than the alignment itself.
The same selection drives the perception radar above and the breakdown table below.
Every one of the 12 measured axes, grouped by the action this run wrote on it. One state at a time: the tab says how many axes sit behind it.
The model places the brand in the bottom third of the peer set on this axis: close to nothing is held here.
A middling read, between the 30th and 70th percentile of the peer set: the model has the idea but holds it loosely.
What to do about these axes1 action
Publish reference documentation that answers a task end to end
Runnable examples, error tables and migration notes on pages that are crawlable without a login, rather than an API reference that assumes the reader already knows the product.
The model places the brand in the top third of the peer set on this axis. It is held.
- Target(2)
- Strengthen(4)
- Already held(6)
- How firmly the axis is held
The model's sentiment towards Stripe
Twelve buying dimensions, each read directly from the model's activation space rather than asked in natural language. Green means the model reads Stripe above the cohort on that dimension, red below it. Their mean, scaled to the same clamp the charts here draw against, is S in the score: +0.76.
The same selection drives the concept action network and the breakdown table below.
How this is measured
Sentiment is read as a direction inside the model's own activation space: contrast pairs of positive and negative words fit a single axis, and the brand's activation is projected onto it, following Tigges et al. (2023). The grey shape overlaid on the radar is Adyen, the highest-scoring brand in payments.
Trajectory: the model reads Stripe above the cohort here, above the payments peer median. This is the claim the model is most prepared to support unprompted.
Ease of use: the model reads Stripe below the cohort here, still above the payments peer median. When an assistant hedges about Stripe, this dimension is the likeliest reason.
What to do about this1 item
- The model reads Stripe below its named peers on documentation.Close this
Publish reference documentation that answers a task end to end
Runnable examples, error tables and migration notes on pages that are crawlable without a login, rather than an API reference that assumes the reader already knows the product.
https://stripe.com/docs
These actions were planned from this audit’s own measurements: the surfaces the crawl found, what your own pages say in their first sentences, and where the twelve dimensions place you against the competitors you named. They are readings of what the model has absorbed, not interventions anybody has run, so none of them promises a movement in a score.
Where you sit, and what separates you
Stripe and its payments peers, placed by a 2D projection of the model's representation space. The map answers two questions: where the model puts Stripe, and which brands it is most likely to be confused with. What separates it from the brands at the core of payments, and what closing that would require, is beside it in units the projection cannot distort.
- Stripe
- Category peer
- Middle of payments
- ModelRank
- Sentiment, positive to negative
Scroll or pinch to zoom, drag to pan. A 2D projection, so read it for who sits next to whom, not for how far apart they are.
The model places Stripe next to Adyen, Braintreepayments, Squareup, and those are the brands it is most likely to be confused with. At the core of payments it holds Adyen, Authorize, Checkout, alongside Stripe: a reference set named rather than averaged into a centroid.
Stripe scores 79, at or above the 62 at the core of payments. There is no gap to close here; the parts below are where the remaining differences are.
Stripe 84.6 against core 43.1, of 100. That is +26.6 points of the score difference.
Stripe 50.0 against core 80.0, of 100. That is -12.8 points of the score difference.
Say what the product is in the opening sentence of every page you own
The homepage, the docs index, the help centre landing page and the pricing page should each name the category in the same words.
Stripe 0.76 against core 0.44, on -1 to +1. That is +3.4 points of the score difference.
Stripe 1.00 against core 1.00, on 0 to 1. Reported alongside the gap rather than inside it, so the three that do sum still sum exactly.
How to read this gapPoints of ModelRank, not projection distance; three components sum, accuracy sits alongside
The headline is a difference in ModelRank score points, because the map beside it is a 2D projection whose distances are not meaningful and cannot honestly be measured off. The reference set is three brands by name rather than an average, so every number here can be checked against those brands. Memory depth, category co-occurrence and sentiment valence sum to the score gap exactly; factual accuracy multiplies the verified score rather than the free one, so it is reported alongside the sum and contributes nothing to the number this card leads with.
These are associations measured across the cohort, not effects anyone has intervened to produce, and none of them promises a score movement. One of them is a warning rather than a lever: social prominence is associated with being named and with nothing about being understood, so volume there can raise recall while leaving the category binding and the read exactly where they are.
Does the model have its facts right
The model correctly binds Stripe to payments, preferring it over four length-matched distractor categories. With a margin of 2.64 over the runner-up, the binding is held firmly.
This is A in the score, read as 1.00 on 0 to 1: zero for a deviated or unaddressed binding, and rising with the margin of a consistent one until it reaches 1.00 at a margin of 0.83.
Expand the exact test we ran
We score five complete sentences of the form “Stripe is a payments platform.” and rank them by mean per-token log probability: the true category against four distractors drawn from the cohort's 333 categories. Distractors are matched to the true label's length, so a short or common category name cannot win on fluency alone. Nothing is generated and nothing is asked in natural language, so the result cannot be shaped by how the model chooses to phrase an answer.
The margin of 2.64 is the gap between the winner and the runner-up. Above 0.83, the median among correct bindings in this cohort, we call a binding firmly held; below it, weakly held.
Pipedrive
AttioCompetitive breakdown
Stripe against its payments peers: the composite, the assessment the same rule reaches everywhere on this page, whether the model names each brand at all, whether it files each one under the right category, and how it reads them. Same measurements, same model, ordered by ModelRank. Select any competitor's name to compare Stripe against it, which drives the radar, the dimension bars and the concept comparison above.
| Brand | ModelRank | Assessment | Recall | Category | Sentiment |
|---|---|---|---|---|---|
| StripeYou | 85 | Established | +1.90 | ||
| 43 | Established | +1.09 | |||
| 38 | Absent | +0.45 | |||
| 38 | Absent | +1.32 | |||
| 43 | Established | +2.34 | |||
| 38 | Absent | +0.88 |
Explore our other tools
More free tools to help optimize your content for AI search.
Reddit threads finder
Find threads cited in ChatGPT and Gemini for AI visibility and AEO
Try itAEO content evaluator
Grade your content for AI search optimization and citation readiness
Try itHeadline optimizer
Generate and rank headline variants for AI search and SEO
Try itLLM eval calculator
Calculate optimal sample sizes for AI evaluation experiments
Try itAI Assist Widget
Add AI buttons to any website to boost AEO and brand citations
Try itAgentic browsing checker
Run the Lighthouse agentic-browsing audit: WebMCP tools, schemas, llms.txt, CLS
Try it