Static, fully crawlable version of the Cy▲ FAQ. For the interactive experience, visit creativitydelta.com/faq.
Cy▲ is an AI-powered auditing solution at the intersection of marketing and localization. Designed for brands and agencies, it evaluates the strength and relevance of marketing content, from product descriptions to social media.
In the AI era, spotting typos is no longer the challenge. The real battlefield is impact. Cy▲ measures whether your content is actually doing its job: conveying your brand voice and driving sales based on your specific positioning.
Using a proprietary framework of 6 compartments and 161 data points, Cy▲ assesses your content via a structured scoring grid. By benchmarking several pieces of content against your "Gold Standard" text, the system measures the Creativity Delta: the precise gap between where your brand voice is and where it needs to be.
Identify exactly where your content loses its edge. Stop guessing and start measuring your content's performance with a granular Creativity Score.
Sensitive content is processed transiently, never retained. Texts and your own API-key credentials travel over HTTPS to stateless Edge Functions, are held only in request memory long enough to call the selected provider, then are discarded. They are not written to the Cy▲ database or to application logs.
When you enter your own API key (e.g., OpenAI, Anthropic), it lives in your browser session and, for the duration of a call, in the transient memory of the Edge Function that forwards it to your provider. It is never written to a database, a cookie or a log. Closing the tab clears it from your browser. Once the request reaches your provider, its handling follows that provider's own API terms.
Your audited content is volatile: source text is never stored. Only the abstract audit output — numerical indicator results and calculated metrics — may be persisted, and it contains no original content. Think of it like storing "78/100" without storing the essay that earned it.
Source texts, generated variants and pre-check results are not persisted. They live in your browser session only and disappear after a hard reload.
Prompt instructions you save are persisted exclusively in your browser's local storage, not in the Cy▲ database. When you select one for generation, its instruction is transmitted transiently with that request and discarded with the other request data. Clearing your browser site data permanently deletes the saved local copy.
Fair-use accounting for the shared GPT-5 Nano key records only your user id, the UTC day, an item count and timestamps. No text, prompt, output or provider key is involved, and this metadata is automatically retained for no longer than seven rolling days.
Each analysis spawns 18 isolated API calls (one per scoring category). This prevents cross-contamination between texts, but also means no single call ever sees the full picture — compartmentalization doubles as a privacy safeguard.
Even the downloadable PDF report is generated entirely in your browser. The document never transits through a server.
Cy▲ audits your text across six strategic pillars that play a critical role in converting your customer base:
This high level of granularity allows you to immediately see where your content is strong and where it loses its impact compared to your selected benchmark text.
First, you select your texts. Cy▲ analyzes each one in complete isolation, and scores are only compared at the very end. This "compartmentalized" design prevents "evaluation pollution" (where a localized copy is unfairly penalized by its source). This allows you to use any text as a benchmark, such as a high-performing product page or a viral post, instead of being limited to a traditional source text.
Once you start the audit, Cy▲ sends a separate API call for every text. Technically, Cy▲ isn't just one auditor; it is a specialized expert that "clones" itself as many times as necessary to provide stable, independent evaluations for each piece of content.
The LLM detects the presence or absence of 161 linguistic traits across 18 categories. To ensure total objectivity, the LLM is never told which traits are "good" or "bad" for your brand (it simply identifies what is there). To prevent AI "leniency" (the mushroom effect), the framework is full of contradictory questions. By forcing the LLM to choose between opposing traits, like asking if a register is both "Formal" and "Casual," we ensure it stays grounded in the text.
While the first results appear, Cy▲ runs a second simultaneous audit in the background. The final score for each indicator is the arithmetic average of both passes (yielding 0, 0.5, or 1), applied uniformly across all questions regardless of their weight or compartment. This averaging smooths out LLM volatility, and any inter-pass disagreements are surfaced in your Trust Index.
After detection, the scoring happens. Unlike traditional systems, Cy▲ separates detection from weighting. The LLM acts only as a detector (outputting 1s and 0s), while the final score is calculated by an external script using a fixed Scoring Grid. Because the LLM does not know the multipliers, it cannot "game the system" or be overly lenient.
Finally, the results are refined by your chosen brand DNA. This architecture allows us to reward or penalize the same trait depending on the grid. For instance, an "Elevated Register" is rewarded in the Premium Luxe grid but penalized in a Lifestyle context. The Standard grid provides a balanced benchmark for all marketing types.
Select your LLM (GPT-5.4 recommended), enter your API key, and choose your brand profile (Standard, Luxe, Lifestyle, or Sport) to set your scoring grid.
Paste, drop, or upload up to seven texts. Designate Text A as your benchmark for the Creativity Delta calculation.
Indicate your content type and click "Audit my text(s)." Cy▲ automatically performs a dual-pass audit and averages both passes per indicator to deliver maximum precision.
View individual Creativity Scores and the Creativity Delta vs. Text A. Generate pillar-specific feedback or export your data via CSV or PDF.
A Full Audit is the default: all six compartments are evaluated across the complete 161-indicator framework, and every metric is available.
A Partial Audit evaluates only the compartments you select. The indicators of the unselected compartments are never sent to the model, so a partial run makes fewer calls and takes less time and provider budget than a full run. The reduction depends on which compartments you keep, because they do not contain the same number of categories.
A Partial Creativity Score keeps the same weighted logic as a full score, but it only sums the compartments that were evaluated. It is therefore comparable only between texts and runs using the identical compartment selection and the identical scoring grid — never against a Full Audit or a different partial scope.
The Trust Index, the Foundation Score and the Profile Adherence score all depend on the complete framework. They require a Full Audit and are unavailable — not zero — in a Partial Audit, in the app and in every export.
The Translation Station is Cy▲'s pre-audit workspace. It lets you start from one source text, create controlled translation or localisation variants in Text B–G, run a fluency pre-check, and decide which texts enter the audit.
Cy▲ does not only show you what changed after translation. The Station lets you create the variants yourself, keep the comparison variables under control, and then measure their impact through the same 161-indicator audit. Translation and pre-check remain separate from the audit engine and never alter its scoring methodology.
An occupied destination slot requires an explicit overwrite confirmation, and the source slot itself can never be overwritten.
Yes. Prompt testing is one of the Station's core use cases. Keep the same source text, generate several variants with different prompts, then audit the outputs side by side to see which instruction preserves or improves the intended marketing impact.
The Station includes three starting points: Faithful translation, Marketing localisation and Transcreation. You can also create your own reusable prompt instructions. Custom prompts are stored locally in your browser, not in the Cy▲ database.
You can generate several target languages, reuse one language with different prompts, or specify a custom language or locale through Other…. Every result is written to the exact Text B–G slot you selected, so the comparison identity remains stable from generation to audit and export.
This makes the Station useful for translation prompt QA, localisation experiments, human-versus-machine comparisons and controlled transcreation tests.
The pre-check looks only for fluency errors before the creative audit: grammar, spelling, agreement, consistency in terminology, logic and locale conventions.
It does not comment on style, brand voice, persuasiveness or creative strength. Those dimensions belong to the full Cy▲ audit. A text can be error-free and creatively weak, or contain a correctable fluency issue while still expressing a strong brand voice.
When an issue is detected, the Station displays its category, the relevant excerpt and a proposed correction. A malformed or contradictory model response is shown as an error; it is never silently converted into "No issue found".
No. The workflow is flagged, user decides.
The pre-check is a decision aid, not a gatekeeper. You remain responsible for deciding whether a flagged issue is relevant in context.
In Generate, each output has its own LLM selector, so you can run several models side by side in a single batch, each with your own session-only key for that provider. The Fluency pre-check uses the model and API key currently selected in the dashboard header. There is no silent model substitution.
Each generated output requires one LLM call. Each text included in a fluency pre-check requires one additional LLM call. Batched outputs and checks are processed in parallel where possible, but each item remains a separate provider operation.
The shared Nano key is provided for feature discovery and testing only. Translation Station usage is limited to 30 items per user per UTC day across translation and pre-check, with a maximum of 40,000 source characters per request. The allowance resets at 00:00 UTC.
When you use your own provider key, Cy▲ does not apply this daily Station quota. Your provider charges the calls directly according to its own pricing and terms. Professional audits should use the recommended high-capability model, not Nano.
No. It is a lightweight pre-audit prompt lab and variant generator, not a replacement for a production translation environment.
Creates controlled translation/localisation variants, lets you compare prompts or languages, checks basic fluency and prepares an audit pool.
It does not provide translation memories, termbases, bilingual file alignment, package handling, segment-level review or project delivery workflows. Use your CAT/TMS for production translation management and Cy▲ to test instructions, triage variants and measure creative impact.
Yes. The Cy▲ Public API v1 lets you run full audits from n8n, Make, Zapier, a backend job or any HTTP client. One request sends your texts and audit settings; the synchronous JSON response returns the Creativity Score, compartment scores, Trust Index, Foundation Score, Profile Adherence and the 161-indicator result metadata.
The current public endpoint automates the full Cy▲ audit. Translation Station generation and fluency pre-check are not exposed as public API operations at this stage.
Your Cy▲ API key authenticates your account, profile, scoring-grid access and fair-use allowance. Your LLM provider key authorizes the selected model and is supplied separately in each request. It is forwarded transiently to the provider and is never stored or logged. A new Cy▲ API key is displayed only once; the platform retains only a one-way fingerprint.
The API supports idempotency keys for safe retries and currently allows 10 audit requests per minute and 500 audits per month. A ready-made n8n workflow is available from the API area of the dashboard.
API texts, provider credentials and returned scores are processed in memory only and are not persisted. Provider-side handling remains governed by the selected provider's API terms.
Read the full API documentation →
You can choose between two modes based on your needs:
Use our shared key with GPT-5 Nano to explore the interface for free. This is for discovery only; the model is lightweight and lacks the precision required for a professional brand voice audit.
To unlock the full Cy▲ protocol, we recommend using your own key. This grants you access to GPT-5.4, our benchmarked standard for reliability and accuracy.
You pay a one-off fee for the Cy▲ platform and pay the LLM provider directly for your tokens. This consumption-based model ensures you only pay for what you use. Low-volume users never subsidize heavy users.
By using your own key, you are never restricted to cheaper, lower-performing models. You maintain total independence and access to the highest-tier models (GPT-5.4) for uncompromising audit results.
There are no hidden markups or intermediary fees on AI processing. You pay the market rate for the "job done" directly to the source.
The Trust Index is a reliability metric designed to verify the quality of the AI's audit. It determines whether the LLM evaluated the content with high confidence or if the results were impacted by ambiguities in the source text or fluctuations in the model's reasoning.
The index is calculated using three technical components:
This monitors the consistency of the AI's judgment between audit passes. Because marketing copy sometimes contains "borderline" traits, the LLM may drift between a Yes and No response. The Trust Index identifies these shifts; a higher frequency of changed answers indicates lower confidence, resulting in a trust penalty.
The Cy▲ framework includes intentionally contradictory questions within and across categories to verify the AI's honesty. If the LLM provides mutually exclusive scores (such as labeling a text as both "Restrained" and "Bold"), the system identifies these as strong conflicts or tensions, lowering the Trust Index.
Every piece of content has a natural "signal capacity" based on its format. For example, a social media post is expected to have a different density of linguistic traits than hero copy. The Trust Index checks if the signals detected by the LLM fit within a relevant range for your content type, proving the AI did not over-detect traits that aren't truly there.
The Foundation Score represents the structural integrity and baseline viability of your content. While the Creativity Score measures creative impact, the Foundation Score evaluates whether your text meets the essential quality standards required for professional marketing.
Throughout the audit process, the AI encounters a series of "quality thresholds" embedded within every thematic domain. These thresholds are designed as forensic checkpoints to identify fundamental flaws, such as rhythmic disruption, syntactic friction, or tone inconsistencies, that would otherwise undermine the effectiveness of the copy.
This score is expressed as a percentage, reflecting how many of these critical thresholds the text cleared. A 100% Foundation Score confirms that the structural baseline was met across the entire framework, ensuring your content is built on a solid professional footing before creative nuances are measured.
The Profile Adherence score is a high-level strategic index that measures your content's alignment with a specific brand voice. It is the most precise way to determine if your text truly "sounds" like the profile you have selected (Luxe, Sport, or Lifestyle).
The score focuses exclusively on the most critical indicators within our multiplier grid, the "make or break" traits of a brand profile. Mathematically, it isolates questions with the highest rewards and the heaviest penalties.
To achieve 100% adherence, the auditor must:
Available on Premium profiles, this percentage allows you to see at a single glance if a text captures the essence of an "artefact" brand. For users with Tailored Profiles, this index is automatically customized to answer your most important question: "Does this content respect my unique brand DNA?"
To get the most out of the Cy▲ framework, choosing the right reasoning engine is key. Your choice depends on whether you are testing the interface or conducting a professional brand audit.
This lightweight model is provided for free to help you explore the app's features and workflow. It is test-only: its limited stability and reasoning depth mean audit results cannot be trusted and it should never be used for professional scoring decisions.
For actionable KPIs, we recommend using your own API key with one of the following models:
The gold standard for Cy▲. It offers 95% average stability across all profiles (Luxe, Sport, Lifestyle). It is an exacting judge, ensuring that your scores are based on high-standard, consistent logic.
Perfectly stable but tends toward leniency. It may overlook subtle friction or quality issues that a more rigorous model would flag, or over-detect linguistic traits whose real impact is negligible.
Currently unsuitable for settling KPI scores. With under 80% stability, the "drift" between audits is too significant for precise benchmarking.
Its extreme rigor led us to halt further investigation into its stability for standard auditing purposes.
Currently, saving analyses directly within the web app is not supported. We are, however, developing a feature that will allow clients to create custom dashboards within the platform.
In the meantime, you can export your audit results as a .csv file to build your own custom dashboards.
Yes. You can export your audit results in two formats directly from the dashboard:
This format is ideal for deep-data analysis or internal tracking. The export includes:
The PDF report provides a comprehensive overview designed for presentations and quick reviews. It includes: