Turn your SEO process into AI-powered tools

18 June 2026 at 18:00

Turn your SEO process into AI-powered tools

Ask ChatGPT or Gemini to “review my on-page SEO,” and you’ll get a perfectly reasonable answer.

Reasonable. Generic. Boring. Uninspired. And almost identical to the answer your competitors get when they ask the same question.

That’s the problem with AI out of the box. It’s a generalist. It knows a little about everything and nothing about you — your business, your customers, your market, or the way you do SEO. The questions are loosely framed and inevitably come back with general answers.

The good news is that’s also the opportunity. The same tools that produce generic answers can become specialist assistants that encode your knowledge, process, and standards. No code required.

Building one is simpler than most people think. With tools like GPTs, Gems, and Claude Projects, you can package your SEO process into a reusable assistant that helps identify opportunities, automate repetitive tasks, and apply your expertise consistently.

Why generic AI gives generic answers

You don’t need a computer science degree here, but a basic understanding of how AI works helps explain the benefits of this approach.

Large language models are prediction engines. They’ve been trained on a huge slice of the internet and human knowledge, and when you ask a question, they predict the most plausible response based on everything they’ve seen.

In other words, by default, you get something close to the internet’s average opinion on a topic.

For SEO, the internet’s average opinion is … fine. It’s the same advice repeated across a million articles. Check your title tags. Improve your content. Build some links. Blah.

What the model doesn’t know is anything about your specific situation:

Your business, services, and commercial priorities.
Your marketplace and competitors.
Your customers and the problems they’re trying to solve.
Your way of working — the checklists, thresholds, and judgment calls you’ve refined over years.

The output is only as contextual as the input. Give it nothing, and you get the average. Give it your knowledge, and you get something far more useful.

It’s a computing problem as old as computing itself: garbage in, garbage out (GIGO).

Be the brand AI recommends.

See where your brand appears in AI search, where competitors are winning, and what it takes to become the answer AI recommends.

See your AI visibility

From generalist to specialist

There are a few ways to add that missing context, in increasing order of effort:

Better prompts: Include context in your question: who you are, what the business does, who the customer is, and what good looks like. This works, but you end up pasting the same 500-word preamble into every chat. It’s tedious and easy to skip when you’re busy, which impacts the quality of the output.
Custom instructions and knowledge files: Most AI platforms now let you save a set of standing instructions and upload reference documents. The AI reads these every time, so you set the context once and it persists.
Simple AI apps: Package those instructions and documents into a named, reusable tool with a specific job. This is where GPTs and Gems come in.
Actual software: Use AI coding tools to build real scripts and applications when you need automation beyond a chat interface.

The great thing is that the jump from a “big prompt” to a “simple app” is smaller than it sounds.

The skill is the same: clearly describing the job, the process, and the standards. If you can write a good brief for a junior team member, or a standard operating procedure (SOP), you can absolutely build one of these.

This is an important point because most people assume this is far more complicated than it is, and that assumption is holding them back. You don’t need to be a developer to do this.

The development of custom tools is no longer a heavily technical job. It’s becoming more of a creative endeavor enabled by these new AI tools and the simple, descriptive way of building apps.

If you can document your process, you can build an AI app.

The platforms: GPTs, Gems, Claude, and Replit

A quick tour of the main options for building simple AI apps:

GPTs (ChatGPT): Custom versions of ChatGPT with their own instructions, knowledge files, and capabilities. They’re shareable via the GPT Store, which is handy if you want to publish a tool for clients or your audience.
Gems (Gemini): Google’s equivalent. Custom versions of Gemini with instructions and knowledge files, with the obvious appeal for SEOs living in the Google ecosystem alongside Search Console, Analytics, Drive, and Sheets.
Claude Projects (Claude): Anthropic’s take. Project-level instructions and knowledge with a large context window, so it can hold a lot of your documentation in mind at once. My personal favorite at the moment.
Replit: A browser-based platform where you describe an app in plain English, and AI builds and deploys actual working software. Use this when a chat interface isn’t enough and you want a real tool with a real interface processing real data.
Claude Code: An agentic coding tool from Anthropic where you delegate coding tasks in plain language, and it writes, runs, and fixes the code. It’s brilliant for building scripts that crunch large exports — say, processing a 100,000-row Search Console export that would choke a chat window.

For most SEO and marketing professionals dealing with day-to-day optimization work, the sweet spot is the first tier: GPTs, Gems, or Claude Projects. They take minutes to build, require no code, and capture 80% of the value.

I’ll use Gemini Gems for the worked example below, as it’s the closest to home for those of us who live in Google’s world. The principles transfer directly to GPTs and Claude, and if you want to build something a little more advanced, have a play with Replit.

*Google Gemini interface (on the Gems page)*

Why not use existing SEO tools?

Standard SEO tools are brilliant at what they do — crawling, rank tracking, and link data. I use them every day. But they share a weakness: They’re generic by design, while your business is totally unique (or at least it should be). They have to work for every business in every industry, so they can’t know what matters to you. Everyone sees the same scores, the same recommendations, and the same “issues,” many of which don’t matter for your situation.

The tools are also largely focused on analysis and opportunity. The kinds of tools you can build with AI are more focused on the actual work.

Vanilla AI has the same problem from a different direction. Hugely capable, zero context.

The strength of building your own simple AI tools is personalization:

Your business: The AI knows your services, priorities, and commercial goals.
Your marketplace: It understands your competitors, customers, and niche.
Your knowledge: It applies your process — the way you’ve learned to do this work over the years — rather than the internet’s average.

That last point is the big one. After 30 years of doing this, my honest take is that the value isn’t the AI. The value is the knowledge and process you encode into it. Your experience is what matters — the AI is just your superpower.

What should you automate?

A simple rule: Automate repetitive tasks. Good candidates are tasks that are:

Repetitive: You do them the same way, over and over.
Process-driven: You could write the steps down for a junior team member to follow.
Data-heavy: They involve staring at exports and spotting patterns — exactly what machines are good at and humans get bored with and subsequently do poorly.

Reviewing Search Console data ticks all three boxes. So do first-pass on-page reviews, log file triage, internal link analysis, and monthly reporting prep.

What you don’t automate is judgment: strategy, prioritization against business goals, and the final call on what actually ships. The AI does the legwork and surfaces the candidates. You decide.

Example: Search Console quick-wins Gem

Let’s build a simple tool to help you mine Google Search Console for content ideas and easy wins.

I wrote “How to unlock easy wins in Google Search Console” two years ago, covering the creaky old human way of doing it. Let’s automate it to free up time for the really valuable creative work.

Note: This is a purposely simple example that’s ideal for AI and automation because the task is repetitive and the data is free.

Step 1: Define the job

Write one sentence describing what the tool does:

“Review Google Search Console performance data and identify prioritized quick-win opportunities, with specific recommended actions for each.”

Simple enough.

Step 2: Document your process

This is the important bit, and it’s where you have to think about the process.

What do you actually do here? What process do you follow? What easy wins and opportunities are you looking for?

Striking-distance keywords: Queries ranking just off page one (or just off the top positions) with meaningful impressions. Small improvements here can have an outsized impact.
High impressions, low CTR: You’re visible but not winning the click — usually a title and meta description problem, or a SERP feature is eating your lunch.
Declining queries and pages: Anything trending down versus the previous period that deserves attention before it becomes a problem.
Query-page mismatches: Queries landing on the wrong page, or multiple pages competing for the same query.
Unexpected queries: Things you rank for accidentally that hint at content opportunities.

For each of these, also note the thresholds and judgment calls. What counts as “meaningful impressions” — 100? 500? What CTR is “low” for position 3 versus position 8?

This is your experience being made explicit, possibly for the first time.

Step 3: Write the Gem instructions

Now open Gemini, create a new Gem, and translate that process into instructions. A solid structure is:

Role: Who the Gem is.
Task: What it does with the data it’s given.
Process: The steps, checks, and thresholds — your documented process from Step 2.
Output: The exact format you want back.
Guardrails: What it should never do.

Here’s an abridged example to adapt:

Role: You are an experienced SEO analyst. You are methodical, skeptical, and prioritize commercial impact over vanity metrics.

Task: I will provide an export of Google Search Console performance data (queries and/or pages, with clicks, impressions, CTR, and position). Review it and identify quick-win opportunities.

Process: Check for, in priority order:

Striking-distance queries — average position 5–15 with 100+ impressions.

High-impression, low-CTR queries — flag where CTR is significantly below what you’d expect for that position.

Pages or queries declining versus the comparison period.

Multiple pages ranking for the same query.

Output: A prioritized table with opportunity, query/page, current metrics, recommended action, and expected impact (high/medium/low). Maximum 15 rows. Quality over quantity.

Below the table, provide a short plain-English summary of the three actions I should take first.

Guardrails: Only use the data provided. Never invent queries, pages, or metrics. If the data is insufficient to assess something, say so. Ask clarifying questions if the export format is unclear.

That guardrails section matters more than people realize. “Only use the data provided” is your main defense against the AI confidently inventing things.

Get the newsletter search marketers rely on.

See terms.

Step 4: Add knowledge files

Gems can reference uploaded knowledge files. This is where you fine-tune things and add depth without bloating the instructions.

Examples include:

Your on-page optimization checklist (for when the Gem recommends title or content changes).
Your title and meta description guidelines, so suggested rewrites follow your standards.
A short brand and business context document — who the client is, what they sell, and which products or services are commercial priorities.

This lets the Gem prioritize opportunities that matter, not just opportunities that exist. That’s especially important when reviewing Search Console data, as most sites show up for a wide range of searches that aren’t aligned with the client’s core goals.

Step 5: Save it

It really is that simple. Hit save, and you’ve created an AI app.

Step 6: Feed it data and test

Export your performance data from Search Console (Performance report > Export, or via the API or Sheets if you want more rows), then start a chat with your Gem and upload the file.

Browse to the Performance report and click Export in the upper-right corner. In this example, I use Google Sheets to keep everything in the Google ecosystem.

Then upload the file and ask for the output you want.

The Gem’s output — a prioritized quick-wins table for a real site

If at first you don’t succeed: The first output here wasn’t terribly useful for this site.

The recommendations didn’t align with the client’s goals. I had to revisit my third knowledge file regarding the business’s commercial goals and priorities.

After refining that document and running the analysis again, the suggestions became much more useful.

Step 7: Iterate like you would with a junior team member

The first version will get things wrong. That’s expected, and it’s actually the useful part.

A bad recommendation is a way to identify what could be improved. Whatever your answer is, that’s a rule that was missing from the instructions. Add it, and the Gem gets a little closer to working the way you do.

Treat it like a new team member. Review its work, correct it, and update the brief. After a few rounds, you’ll have something that delivers a genuinely useful first pass in seconds — and a documented process that’s valuable in its own right.

A note of caution

Some honesty before you let this loose on client work:

AI gets things wrong: Confidently. Always verify recommendations against the actual data before acting, and never let AI output go straight to a client without review.
Mind the data: GSC exports are business data. Check the privacy and data settings on whatever platform you use, especially when client information is involved, and make sure your approach aligns with any agreements you have in place.
It’s a first pass, not a final answer: The tool surfaces candidates. You supply the judgment. The moment you stop checking is the moment you make a mistake.

More simple SEO tools to build

Once you’ve built one, the pattern repeats. Same recipe — role, task, process, output, guardrails, and knowledge files — different job.

Any manual task you do repeatedly is a good candidate for this type of tooling. Examples include:

Keyword research assistant. Feed it seed terms and keyword exports. It clusters by intent and maps keywords to your site structure using your intent categories and customer personas.
On-page optimization reviewer. Paste a URL’s content and target query. It reviews the page against your checklist and suggests improvements in your preferred style.
Technical SEO triage. Feed it crawl exports. It prioritizes issues based on actual impact for your site rather than default tool severity scores.
Link opportunity finder. Feed it competitor backlink exports. It identifies realistic, relevant prospects based on your criteria and drafts outreach angles.
Content strategist. Load it with your personas and content strategy frameworks. It generates briefs and ideas anchored to real customer problems rather than generic topics.
Analytics insight reviewer. Feed it GA4 exports. It summarizes what changed, why it might have changed, and what’s worth investigating in plain English.
Search Console opportunity finder. The example we just built, easily extended into variants for content decay, cannibalization, or indexing reviews.

Each of these is an afternoon’s work.

The constraint isn’t technical. It’s whether you’ve documented your process clearly enough to hand it over. If not, this is a good opportunity to systemize your business and accelerate the work with a simple app.

You can also apply the same approach across digital marketing: building personas, improving your homepage for AI-driven search, comparing SEO, PPC, and AI strategies, or tackling whatever else falls under the modern marketer’s remit.

If AI can’t find you, customers won’t either.

Track your visibility across AI search, uncover missed opportunities, and grow your presence where customers are asking questions.