My proprietary SEO prompts to improve your rankings

Before applying these prompts, check that your site isn't among the mass of sites LLMs never use. I use prompts not to replace my thinking, but to improve it. I never give the AI something I could do better myself. What I hand to the AI is what I'm not able to do, or an area where it would be better than me. Obviously, I never give it the foundation.

AI always helps me create the SERP's originality on a keyword. All my prompts serve that. I have prompts that let me find these content gaps, make the difference on the SERP, and improve my writing, the customer experience and the user experience on my page.

My tips for SEO prompting

Never let the AI write the foundation or the Hn structure

Don't let the AI write your best content. Google expects you to be the SERP's originality, not an "11th generic result".

• I never let the AI define the Hn structure of my content
• I never let the AI write more than 40% of the article
• I leave it the informational parts with little added value

Understand how LLMs work

I use prompts to do what's called reverse engineering, that is, trying to understand how the AI thinks. All my prompts help me better understand what AIs expect and, by extension, what LLMs will expect, because we know today that Google is fond of them.

Aim for the ideal length (150–300 words)

For tasks of medium complexity, keep your prompt between 150 and 300 words. Avoid going over 500 words, because the model's performance tends to drop beyond that threshold because of "bloat", which dilutes the instructions.

How do I use SEO prompts in my processes?

1. Create "Proprietary Prompts"

These are prompts tested and validated for specific use cases. Generic or public prompts won't reach the quality Google expects.

2. Create workflows

SEO is going to get more and more complex. In that context, you'll need to create workflows, that is, take your proprietary prompts and build them into each step of your SEO strategy to optimise it.

The aim is to improve, for example, how you identify search intent, spot content gaps, and add either expert content or news facts. AI has to be implemented smoothly, with a clear objective: improve the text at every step, from writing to updating.

You have to understand each step, be able to analyse each prompt's output, spot inconsistencies and, above all, innovate on your topic. For six years I've been hearing that you can't rank without backlinks, that if you don't copy others you won't rank, that those who go full AI will take everything. They were wrong.

I sincerely think we are witnessing a fundamental change. These prompts fit into the methodological framework of the semantic grid. Those who master a proprietary SEO workflow will make the difference. Conversely, those who want to automate everything – for lack of thinking, or overwhelmed by the speed of AI's evolution and the contradictory information – will lag behind.

I'm sharing some of my best prompts for improving my strategies

Prompt for semantic clustering

This prompt is used after applying the workflow for selecting the right intents on the AEO side, to group the resulting list.

This prompt lets you create semantic cohorts from your Google Search Console:

1. Go to GSC
2. Export the data
3. Download the "Queries" page
4. Drag the file into your prompt

Context: You are an SEO expert specialising in behavioural analysis of queries. Your role is to group the keywords of a "Search performance" report (Search Console or PDF/CSV export) into homogeneous cohorts, according to their search intents, volumes, CTR and average positions. These cohorts must let an SEO consultant build a clear content and conversion strategy.

Expected output: Generate a complete, professional segmentation of the report provided, with, for each cohort:

• Cohort name (e.g. "Brand queries", "Transactional products", "Technical comparisons", "Local queries")
• Main intent: Awareness / Consideration / Decision
• Examples of queries taken from the document
• Quantitative insights: average CTR, average position, estimated volume
• Qualitative analysis: type of user, stage in the customer journey, level of intent
• SEO recommendations

Prompt to get a semantic score for your text

ROLE: You are a semantic ranking engine (Semantic Ranker) simulating a Transformer architecture (MiniLM / BERT-like). You do not write. You evaluate a text by calculating a Grounding Score (Factual & Vector Anchoring) that simulates a cosine distance between an ideal expected vector and a candidate text.

OBJECTIVE: Evaluate whether a [CANDIDATE TEXT] deserves Position 0 (Featured Snippet / SGE) for a [USER QUERY], by measuring:

• Its factuality
• Its completeness against the intent
• Its structure
• Its useful information density

SCORING ALGORITHM:

• VARIABLE A1, Named Entities (Weight: 20%): Institutional entities, proper nouns, technical industry acronyms
• VARIABLE A2, Quantitative Precision (Weight: 20%): Prices with figures, currencies, durations, deadlines, volumes
• VARIABLE B, Structural Alignment (Weight: 25%): Structure aligned with the query
• VARIABLE C, Expert Vector Vocabulary (Weight: 20%): Industry semantic proximity
• VARIABLE D, Signal / Noise Ratio (Weight: 15%): % of strictly useful sentences

Content critique prompt

Role: You are a content strategist and literary critic specialising in deconstructing dominant messages and identifying untapped narrative themes.

Objective: Analyse the following text, identify its structural or thematic weaknesses, determine a content gap (a missing or unexplored angle on this subject), and generate at least three radically different angles of treatment.

Phase 1: Deconstructing the Dominant Assumption

• The Fundamental Assumption to Challenge
• Analysis of Internal Tensions
• The Expertise Cliché to Avoid

Phase 2: Detecting the Deep Content Gap

• The Uncomfortable Theme
• The Value Inversion
• Micro vs. Macro Scale

Phase 3: Generating Expert, Counter-Intuitive Angles

Generate three clear, punchy angle proposals, each with an impactful title + thesis.

A few questions on prompting and SEO

How do you avoid hallucinations?

Without context, AI hallucinates. To avoid getting false information, the AI needs a "context window". The recommended method is to create a "themed bot" (custom AI) to give it the scope of its field of work. This bot must be fed with exclusive content, semantics, knowledge graph and the client's data.

Does Google penalise content generated with prompts?

Even though you can read everywhere that it doesn't, the answer is much more nuanced because rules exist (Quality raters):

• Google doesn't want generic content, even if it's good quality
• Google wants the content to be at least reviewed by an expert, ideally co-written
• A page should only be published if it brings something new to the SERP

Can a good prompt replace an SEO tool?

I hardly use tools any more. Instead, I use prompts to replace SEO tools that I often find obsolete with generative AI. The aim is to improve what exists using contextual or contextualised data, particularly on semantics. Today, AIs are very good at improving a text's semantics, but at the same time they are very bad at writing it. So you have to find the right balance.

What is the best LLM for running my SEO prompts?

I'd say Gemini 3.0, in the Pro version if possible, keeping in mind that LLMs evolve very quickly. So it's important to stay up to date to identify which one gives the best answers at any given moment. What also matters, and I insist on this: whether you use ChatGPT, Perplexity or Claude, always give your prompt context.