AI features

FilaRank is bring-your-own-key. Add a key under FilaRank SEO → SEO Settings (stored encrypted via Crypt), or fall back to env: OPENAI_API_KEY, GEMINI_API_KEY, ANTHROPIC_API_KEY. Nothing runs unless a key is configured. AI actions hide themselves when AiService::isConfigured() is false.

Feature	Where	What it does
Fix with AI	SEO section header, per record	Suggests title, meta description, and slug, shown as a before/after diff before anything is written
Audit fix plan	SEO Health → Technical audit	Turns an audit issue into a numbered remediation plan
Keyword enrichment	Keyword Research	Adds intent, difficulty, and volume estimates to a seed keyword
Competitor gaps	Competitors	Suggests keywords a competitor likely targets that you don’t
Backlink toxicity	Backlinks	Scores a link 0–100 for spamminess

Every call is logged to filarank_ai_logs with provider, model, token counts, and estimated cost, surfaced as a monthly spend figure on the Overview dashboard. Use Test AI connection in Settings to do a real round trip and surface a bad key or unsupported model immediately.

Choosing a model

The model dropdown in SEO Settings is built from config, and that list is the allowlist. Add a model and it becomes selectable, with no code change:

'ai' => [
    'providers' => [
        'openai' => [
            'api_key' => env('OPENAI_API_KEY'),
            'models' => ['gpt-4o-mini', 'gpt-4o', 'your-new-model'],
            'rates' => [
                'default' => ['in' => 0.15, 'out' => 0.60],
                'gpt-4o' => ['in' => 2.50, 'out' => 10.00],
            ],
        ],
    ],
],

The first entry is the default when nothing has been chosen. A saved selection is validated against the list on every read, so removing a model from config quietly falls back to that first entry rather than erroring. rates are USD per 1M tokens and drive the spend estimate. Key a rate by model name to price it exactly; anything without its own entry uses default. Getting this wrong only skews the dashboard figure, never a request.

Requests are built per provider, not per model. The OpenAI provider sends temperature and a JSON response format, so reasoning-style models that reject a custom temperature will fail with a 400. Because publishing the config replaces the package’s ai block wholesale, a published copy will not inherit providers added in later versions.

Using the service in code

use Usamamuneerchaudhary\FilaRank\Ai\AiService;
use Usamamuneerchaudhary\FilaRank\Ai\Providers\FakeProvider;

$suggestion = app(AiService::class)->suggestSeoFix([
    'title' => $post->seo->title,
    'description' => $post->seo->description,
    'focus_keyword' => 'coffee beans',
    'slug' => $post->slug,
    'content' => $post->content,
]);

// Deterministic output for tests (local/testing environments only):
(new AiService)->useProvider(new FakeProvider)->suggestSeoFix([...]);

All provider failures are wrapped in AiRequestException with credentials stripped out, so a bad key or rate limit never leaks a secret into your logs or a notification. The fake provider is refused outside local and testing.

Getting Started

Usage

Analysis

Site health

Research

AI

Choosing a model

Using the service in code

​Choosing a model

​Using the service in code

Choosing a model

Using the service in code