Recently I joined the SEO in 2024 podcast hosted by David Bain, where we talked through what I believe is one of the most practical AI upgrades any SEO agency or consultant can make: building a dedicated prompt library for each client. The conversation focused on why generic AI prompts fall short, and how a structured, client-specific prompt system changes the quality and speed of content production entirely. Below are the key insights from that session, turned into a guide you can start using straight away.
Watch the video: SEO in 2024 — Victoria Olsina on AI Prompt Templates for SEO
What Is a Prompt Library and Why Does Every SEO Client Need One?
A prompt library is a structured collection of AI instructions — one for each content type a client regularly produces. Rather than writing a new prompt from scratch every time you need a LinkedIn post or a product page, you open the library, pick the relevant prompt, and run it.
The difference between a generic prompt and a client-specific one is significant. Generic prompts produce generic output — content that sounds like every other AI-written article. A prompt built around a specific client’s tone of voice, audience, CTAs, and content style produces output that actually sounds like them.
What Goes Into a Prompt Library
The most common content types that benefit from dedicated prompts:
- LinkedIn posts — how the brand speaks to its professional audience, what topics it covers, what tone it uses
- Newsletter intros — one of the most time-consuming content tasks for most clients; a well-built prompt cuts production time dramatically
- Product pages — client-specific CTA language, feature and benefit framing, compliance requirements where relevant
- Blog posts — structure, tone, internal linking approach, keyword density preferences
- Press releases — brand voice, approved terminology, formatting rules
- Guides and reports — length, structure, citation style, audience level
Each of these needs its own prompt. A LinkedIn post is not a product page. Trying to use one instruction set for both produces weak output for both.
How to Build AI Prompt Templates That Actually Work
The quality of the output depends entirely on the quality of the input. This is the most important principle when building prompt libraries — and the one most people skip.
Start With Examples, Not Instructions
The most effective way to prime an AI for a specific client is not to write “act like a compliance officer” or “write in a formal tone.” It is to give the AI examples of content the client actually likes.
Ask the client: what are three to five newsletters, LinkedIn posts, or product pages they consider excellent — either their own or from competitors they admire? Feed those examples directly into the prompt. The AI will infer tone, structure, formality, and style from the examples far more accurately than from a written description alone.
If the client does not have examples, ask them to subscribe to two or three competitor newsletters and send over ones they wish they had written. That is more than enough to get started.
How to Structure the Prompt Itself
A strong prompt for a client library contains:
- Role or context — who the AI is writing as, who the audience is
- Examples — 3-10 real samples of the content type the client approves of
- Tone and style rules — specific instructions about formality, vocabulary, words to avoid
- Output format — length, structure, heading style, CTA placement
- Avoid list — terms, phrases, or topics the client does not want used
Once the prompt is drafted, run it back through the AI and ask: “What is this prompt missing? How would you improve it?” This self-review step consistently surfaces gaps — language preferences, missing context, ambiguous instructions — that only become visible when the model explains what it would need to do better.
Where to Store the Library
The most straightforward storage option is a Google Doc or a Notion page — one document per client, one section per content type. A prompt should be readable, editable, and easy to copy. If you use AIPRM — a Chrome extension that connects to ChatGPT — you can save prompts there and share them directly with clients via a link, without the client ever seeing the raw prompt text.
Iterating and Maintaining the Prompt Library
A prompt library is not finished on the first draft. The first version will get you 70-80% of the way there. Getting to the client’s standard requires iteration.
The Iteration Process
Run the prompt, review the output with the client or a content writer, and collect specific feedback. Not “this doesn’t sound right” but “we never use the word ‘utilise'” or “our CTAs always end with a question.” That specific feedback becomes a new instruction in the prompt.
A reasonable cycle: use the prompt for four to six weeks, collect feedback, refine the prompt, repeat. The prompt improves continuously as long as the feedback loop stays open.
Approved, finalised content should also feed back into the library as additional examples. The more high-quality samples the prompt contains, the more consistent the output becomes over time.
Keeping the Library Current
When a client changes their branding, updates their CTA, launches a new product, or rebrands entirely, the prompt library needs to reflect that. Maintaining it centrally — rather than in scattered individual conversations — means one update propagates across every use.
The Honest Downsides of AI Prompt Templates
The benefits are real, but so are the limitations. Worth being clear with clients about both.
Prompts require time to build properly. Getting a newsletter intro prompt to the point where the client is genuinely happy can take five to ten hours of iteration. That is time worth spending if the newsletter is produced weekly — but the upfront investment needs to be scoped honestly.
AI still hallucinates. Fact-checking is non-negotiable, especially for regulated industries. Solicitors, financial services, compliance-heavy B2B products — errors in this content can have real consequences. Every output needs a human review step before publication.
Repetitive phrasing creeps in. AI tends to reach for the same words and sentence constructions repeatedly. Adding an explicit avoid list to the prompt helps, and a human editor will naturally interrupt the pattern with vocabulary the AI would not choose. That combination — prompt constraints plus human editing — is where the output starts to sound genuinely on-brand.
Automating Beyond the Prompt Library
Once the prompt library is running, the next step is connecting it to automation tools. Make.com (formerly Integromat) and Zapier allow you to set triggers that feed content into an AI module automatically.
A practical example from the interview: HARO (Help A Reporter Out) sends three emails per day, Monday to Friday, each containing dozens of journalist requests. Manually reviewing and responding to every relevant query is extremely time-consuming. An automation that monitors incoming HARO emails, filters for relevant topics (SEO, marketing, AI), and generates a draft response using a pre-built prompt handles the routine work automatically. The human reviews and sends only the drafts that are worth submitting.
This is the logical extension of a prompt library — moving from manual use to triggered automation as the prompts become reliable enough to run without direct supervision.
Conclusion
Building a prompt library is not glamorous work. It is research, iteration, and careful calibration of instructions. But once it is running well, it compounds. Every client interaction produces better output than the last, the human review step becomes lighter, and the time saved on content production gets redirected to higher-value work.
Start with one client, one content type, five examples. Build the prompt, ask the AI to review it, iterate with the client. Then build the next one.
Here’s every single interview of the Majestic podcast series:
- SEO in 2023: The importance of using hybrid models for AI
- SEO in 2024: Build your own prompt libraries for your clients
- SEO in 2025: Improve the output from AI by creating custom GPTs for your clients
- SEO in 2026: Customise AI to automate your SEO processes
Frequently Asked Questions
What is an AI prompt library and how does it differ from a standard ChatGPT prompt?
An AI prompt library is a structured collection of pre-built, client-specific instructions — one for each content type a brand regularly produces. Unlike a standard one-off prompt, a library prompt contains the client’s tone of voice, audience context, examples of approved content, formatting rules, and a list of terms to avoid. The result is consistent, on-brand output without needing to re-explain the client every time. Prompt libraries are stored in Google Docs, Notion, or tools like AIPRM and reused across every content production session.
How do I build AI prompt templates that produce consistent, on-brand content?
Start with examples rather than descriptions. Collect 3-10 pieces of content the client considers excellent — their own or from competitors they admire — and include them directly in the prompt. Add tone rules, formatting requirements, CTA language, and a list of words to avoid. Once drafted, ask the AI to review its own instructions and identify what is missing. Run the prompt, collect specific client feedback, and refine. The prompt improves with every iteration cycle. AI prompt templates built on real examples consistently outperform those built on written descriptions of tone alone.
What content types should I create prompt templates for first?
Prioritise the content types the client produces most frequently and finds most time-consuming. For most B2B clients, this means LinkedIn posts, newsletter intros, and product pages. These three content types require the most brand-specific calibration and are produced on a recurring basis, making the upfront investment in a well-built prompt worthwhile quickly. Blog posts, press releases, and guides are also strong candidates once the foundational prompts are running reliably.
What are the main risks of using AI-generated content with client prompt libraries?
The two most significant risks are hallucination and repetitive phrasing. AI models can generate factually incorrect information — this is especially dangerous in regulated industries like legal, financial services, and compliance-heavy B2B sectors. Every output requires a human fact-checking step before publication. Repetitive phrasing is a separate issue: AI tends to reach for the same vocabulary repeatedly. Adding an explicit avoid list to the prompt and including a human editing step both help interrupt this pattern. Building a prompt library does not eliminate the need for human review — it reduces how much review is needed.
How often should I update a client’s prompt library?
Update it whenever feedback accumulates, the client’s branding changes, new CTAs are introduced, or new products launch. A practical maintenance cycle: use the prompt for four to six weeks, collect specific feedback from the client or content team, refine the prompt, repeat. Approved content should also feed back into the library as additional examples — the more high-quality samples the prompt contains, the more reliable the output becomes. Maintaining the library centrally means one update applies to every future use.
Can prompt libraries be connected to automation tools like Make.com or Zapier?
Yes. Once prompts are reliable enough to run without direct supervision, they can be connected to Make.com or Zapier to trigger content generation automatically. A practical example: an automation that monitors HARO email alerts, filters for relevant journalist queries, and drafts a response using a pre-built prompt — reducing a daily manual task to a review-and-send workflow. OpenAI Assistants supports API-level integration, making it possible to trigger prompt library usage from external systems and push output downstream into a CMS or review tool.
Need help turning ideas like this into working systems?
This is the kind of AI marketing work we do with teams that want repeatable workflows, not one-off experiments. Let’s build yours →
