AI and web audits: how Orilyt uses Claude and GPT for actionable reports
Full transparency on Orilyt's AI architecture: Claude for executive summaries, GPT for SEO suggestions, and why AI doesn't replace your expertise — it makes it visible.
- Orilyt uses two AI providers — Anthropic Claude for executive summaries and OpenAI GPT-4o-mini for SEO suggestions — each in its own role in the report pipeline
- AI doesn't run the tests: Orilyt's 80+ checks are deterministic code. AI interprets the results in human language, in the client's language
- The freelance remains the expert — AI is the translator between technical data and business language. Reports work 100% without AI if it's unavailable
In 2026, AI has made its way into everything: development tools, marketing platforms, CRMs. But in a web audit tool, what does AI actually do? And more importantly, what does it do that you — as a freelance or agency — couldn't do yourself in a few minutes?
Orilyt uses two AI models — Anthropic Claude and OpenAI GPT — not to analyze sites for you, but to do something precise: turn 56 raw technical results into language your client can understand without any prior web knowledge.
This post is a transparent description of that architecture. What Claude and GPT do in Orilyt, how it's implemented, what AI doesn't do — and why that distinction matters for your commercial positioning.
Why integrate AI into a web audit tool?
An Orilyt audit produces 56 results: scores, statuses, technical recommendations, raw data. For a developer or WordPress expert, that table is readable. For the owner of a bakery or a medical practice — your actual client — it's incomprehensible.
A score of 35/100 on SSL security says nothing to someone who doesn't know what a certificate is. A list of 12 red tests doesn't generate a purchase decision. And a non-contextualized technical recommendation — "enable Brotli compression" — doesn't justify a quote.
AI bridges that gap. It reads the audit results as a whole and produces a summary that a non-technical person can understand: what's wrong, why it matters for the business, and what needs to be done. In 3 paragraphs. In the client's language.
That's what AI does in Orilyt. Nothing more — but that "nothing more" is often the difference between a report that gets read and one that gets filed away.
Claude — the executive summary engine
Anthropic Claude generates the hero summary that appears at the top of every Orilyt report. It's the first thing your client sees — and often the only thing they read carefully.
The mechanism is straightforward: when an audit completes, Orilyt sends Claude the full set of 56 results (scores, statuses, raw data) with a precise instruction. Claude returns a 3-paragraph executive briefing: an overall finding, the 2-3 priority issues with their business impact, and an action-oriented closing.
This summary is generated in the report's language — FR, EN, ES or DE — and tuned to context. The tone is professional, actionable and non-alarmist: the goal isn't to scare the client, but to make them want to act.
Technically, summaries are cached in the `audit_ai_texts` table to avoid regenerating on every page load and to control costs. If the Claude API call fails, the report displays normally without the summary block.
GPT — SEO micro-suggestions
OpenAI GPT-4o-mini handles two specific points in the report: meta title and meta description suggestions, and the WordPress mini-summary.
For SEO, GPT analyzes the site's current title and description, the heading structure and main content, then proposes optimized alternatives. These aren't generic "improve your SEO" suggestions — they're concrete proposals based on the real content of the audited page.
The WordPress mini-summary condenses the key findings about the site's WordPress configuration: whether the version is exposed, outdated plugins, security bad practices. It's a section aimed at clients who can't distinguish WordPress from PHP.
GPT-4o-mini was chosen for these tasks for two reasons: its speed (suggestions must appear within a few seconds) and its cost-to-performance ratio for short generation tasks. For the more nuanced executive summaries, Claude is a better fit.
What AI doesn't do in Orilyt
To be transparent about the limits, here is what AI doesn't do in Orilyt — and why that's intentional.
-
AI doesn't run the tests Orilyt's 80+ checks are deterministic code: HTTP calls, header analysis, HTML structure checks, calls to third-party APIs (PageSpeed, Safe Browsing, WPScan). AI receives the results of those tests — it doesn't produce them.
-
AI doesn't decide the scores Every score is calculated by fixed rules in the code. An SSL test scores 0/100 if the certificate is expired, 100/100 if it's valid and well-configured. AI has no influence on that scoring.
-
AI doesn't access client sites Claude and GPT receive only the structured data produced by the audit — never a URL to visit directly, never credentials or access. The isolation is complete.
-
AI doesn't train on your data API calls to Anthropic and OpenAI use standard production models. The audit data transmitted is not used to retrain the models.
-
AI is optional If AI calls fail or if you haven't configured the API keys, the report displays 100% with all tests, scores and recommendations. AI adds value — it's not structurally indispensable.
The technical architecture: async, multilingual, cached
AI calls in Orilyt are designed not to block report rendering. The Claude summary generation happens after the audit is saved, asynchronously when possible.
Caching is central to the architecture. AI-generated texts are stored in the `audit_ai_texts` table with columns `audit_id`, `lang`, `scope` and `payload_json`. A summary generated in FR is not regenerated on every view — it's retrieved from cache.
Multilingual generation is on-demand. If a report is first viewed in FR, the FR summary is generated. If the user switches to EN, the EN summary is generated and cached. All 4 languages (FR, EN, ES, DE) can coexist in the cache for the same audit.
If the API fails (timeout, quota exceeded, server error), the summary block is simply absent from the report — no error message, no broken page. The degradation is silent and graceful.
AI as a business advantage for freelancers
Beyond the technical side, AI in Orilyt has a direct impact on your commercial positioning. Here's what it concretely changes.
The client understands the report without you
When you send an Orilyt report to a client, the Claude executive summary explains in 2 minutes what's wrong and why it's urgent. You no longer need to spend 20 minutes explaining it by email.
You look more professional
A report with a structured executive summary, personalized SEO suggestions and a clear synthesis — that's an agency-level deliverable, not the work of an occasional contractor. AI elevates the perception of your work.
Translation is automatic
Auditing a French site for a client based in Berlin who wants the report in German? Orilyt generates the Claude summary directly in DE. No manual translation, no external tool needed.
Less admin time, more audits
Every hour saved on report writing is an hour available for a new audit. With 20 credits per month on the Solo plan, the goal is for each credit to generate maximum value — and AI contributes directly to that.
AI serving expertise, not replacing it
Integrating AI into Orilyt isn't a marketing argument — it's a response to a real problem: the gap between the technical data of an audit and the action decision of a non-technical client.
Claude and GPT are tools in a pipeline. They don't replace your diagnosis, your client relationship, your ability to prioritize. They do what they do well: generate summary text from structured data, quickly, in multiple languages.
For the freelance or agency using Orilyt, this means more readable reports, better-informed clients, and a stronger professional positioning — without changing how you work.