Someone fires their SEO agency. Buys a Claude subscription for $20 a month. Asks it to write their content strategy, brief their articles, and handle their keyword research. Six months later, nothing is indexed. The site has lost whatever rankings it had. They’re now paying an agency to fix it — and it costs more than the original retainer.
This isn’t a hypothetical. It’s a pattern I’ve watched repeat across the industry since AI tools became cheap and convincing enough to feel like a solution. The tool isn’t the problem. The blind trust in it is.
There is a real place for AI in SEO work. I use it every day. I’ve built systems at MCM that connect directly to Ahrefs and Google Search Console, hold client context in memory files, and run multi-step content briefing workflows that would take twice as long manually. AI makes me faster and, in some ways, more consistent. But the word doing the work in that sentence is me. The AI is infrastructure. I’m still the one doing SEO.
Here’s what that actually looks like — and where the gap between the promise and the reality tends to open up.
What “using AI for SEO” actually means
Before anything else, it’s worth separating two things that keep getting tangled together.
The first is using AI as a tool within your SEO workflow — keyword research, content briefs, technical audits, data analysis, reporting. That’s what this article is about.
The second is optimising your content for AI search platforms — getting cited by ChatGPT, appearing in Perplexity answers, showing up in Gemini responses. That’s a different discipline called GEO (Generative Engine Optimisation), and it has its own logic. If that’s what you’re after, the GEO vs SEO distinction is worth understanding before you dive into tooling.
Most people asking “how do I use AI for SEO?” are asking about the first thing. Most articles answer both at once and muddy the water. Let’s not do that.
The setup no one talks about
Every tool list article tells you which AI tools to buy. None of them tell you what needs to exist before those tools are worth anything. That’s the gap where most people fall down.
Context is everything
Open a fresh ChatGPT or Claude chat window and ask it to help with your SEO. It will produce something. It might even look useful. But it knows nothing about your site, your client, your industry positioning, your tone of voice, your competitors, or the keyword history that explains why your traffic looks the way it does. It’s working from pattern-matching on its training data — which is vast, general, and not about you.
When I build AI systems for client work, the first thing I do is build the context layer. A memory file that holds the client’s site structure, their content history, their performance data, their target audience, and their goals. The AI doesn’t go get this — you have to supply it. But once it’s there, every task runs against that context. The output is coherent with everything that came before it. The AI isn’t guessing.
A bare chat window has none of this. Worse, it can be swayed by previous conversations in the same session, surfacing information or assumptions from an earlier client that have no business being in this one.
Data has to come in — it won’t go get it
This is the one that trips people up most. They expect the AI to pull their rankings, analyse their GSC performance, cross-reference their backlink profile. It doesn’t. At least not without deliberate setup.
The efficiency gain I’ve found most significant in my own workflow is connecting Ahrefs and Google Search Console in one pull — instead of exporting from each platform separately, formatting the data, and then providing it manually. That integration saves real time. But I had to build it. The AI didn’t build it for me.
If you’re using a standard chat tool and manually copying data in, you’re getting some efficiency gain. If you’re providing no data at all and expecting the AI to know what’s happening on your site, you’re not doing SEO — you’re generating plausible-sounding text.
Human approval at every output stage
The most reliable AI SEO system I’ve built is a multi-step content briefing workflow — custom GPT, structured prompts, real keyword and competitor data provided at each stage, and a human (me) approving the output before the next step runs. It’s more efficient than the manual process. It’s also more consistent. But it’s not autonomous.
The approval step isn’t optional. It’s the point where you catch the hallucination, the misread of intent, the fabricated statistic. Take it out and you have a faster way to produce errors at scale.
Where AI genuinely saves time
Given all the above — here’s where it actually earns its place.
First-pass keyword research
With good context and a properly structured prompt, AI can produce a solid first round of keyword candidates. It reduces the time spent manually hunting through tools, surfacing volume data, and clustering by intent. The output isn’t finished — it shouldn’t be — but it’s a useful starting point that cuts the sourcing phase meaningfully.
The key qualifier is “good context.” Ask a bare chat tool for keyword ideas and you’ll get a generic list that could have been written for any business in your category. Feed it your site content, your competitor landscape, and your target audience and it starts producing something worth sorting through.
Data aggregation across platforms
Comparing performance data across multiple inputs in one sitting — instead of platform-hopping and reconciling spreadsheets manually — is one of the clearest efficiency wins I’ve found. A context-aware system can flag patterns across GSC and Ahrefs data that a human reviewing the same data sequentially might miss, or take much longer to surface.
Content briefs and structure
With the right setup, AI produces consistent, structured briefs at speed. Consistent is the operative word — human briefing processes are often inconsistent, which means content quality varies more than it should. AI with good instructions and real data is predictable. That predictability has genuine value at scale.
Technical SEO triage
AI-assisted audits can surface and prioritise issues faster than manual review of a crawl export. Categorising errors, grouping patterns, flagging what’s likely to have the most impact — all of this is faster with AI in the loop.
Note the word: triage. It surfaces. A human still judges what actually matters for this site, at this point in time, with this client’s priorities. That judgment is not something AI can replicate, and it’s where most of the actual SEO value lives.
Where AI goes wrong — and how badly
This is the section the tool list articles skip. They have commercial reasons to skip it. I don’t.
The content trap
Google formalised its scaled content abuse policy in March 2024. The target: sites publishing large volumes of AI-generated content with no meaningful human oversight — thin factual depth, no first-hand experience, no editorial judgment. The enforcement has been significant. One researcher tracked 49,345 websites and found 837 completely deindexed, representing over 20 million monthly organic visits wiped out.
I’ve seen a version of this up close. A client used AI to generate their entire content library — no checks, no editorial layer, no system built around quality control. Zero content indexed. We had to audit the entire site, identify what could be salvaged, rewrite what couldn’t, and rebuild the content strategy from scratch. The cost in time, money, and lost performance was substantially more than it would have cost to do it properly in the first place.
AI content isn’t inherently penalised. Human-edited, genuinely useful content produced with AI assistance performs fine. Mass-produced AI output that adds no value to what already exists is what gets hit. The distinction matters, and it’s one the tool you’re using cannot make for you.
Hallucination as an SEO hazard
LLMs are prediction engines. They predict the next likely word based on patterns in training data. They don’t verify facts. They don’t check whether a URL exists. They don’t know that a statistic they’re citing is three years old or was never real to begin with.
The failure mode I’ve seen most consistently — even in well-constructed setups — is internal linking. Even when I provided a complete list of live URLs for a site, basic ChatGPT chat would still fabricate URLs that didn’t exist. Confidently. In a format that looked exactly right. A junior SEO acting on that output without checking would publish broken links at scale and spend the next week wondering why their crawl reports look broken.
77% of businesses using AI report hallucination as a significant concern (Deloitte). The AI industry spent $12.8 billion on hallucination reduction between 2023 and 2025. The problem is structural — it doesn’t go away because the tool sounds more confident.
Every output that contains a specific URL, statistic, or factual claim needs a human to verify it before it goes anywhere. That step is not negotiable.
The homogenisation problem
When every site uses the same AI tools with the same prompts, the content converges. The same angles, the same structure, the same examples. Google’s systems have become substantially better at identifying derivative content. Consumer research has found trust drops nearly 50% when content feels AI-generated.
The competitive advantage of using AI tools disappears the moment your competitors adopt the same tools. The sites that win are the ones where AI handles the scaffolding and a human with actual expertise adds the perspective that can’t be replicated. Original research. First-hand experience. A specific point of view. The kind of thing you’re reading right now.
How to onboard AI properly
The most useful reframe I’ve found: treat AI like a new hire, not a vending machine.
A new hire needs onboarding. Clear instructions. Access to the right tools and data. Supervised output before they run solo. Feedback loops. You wouldn’t give someone their first day at work, hand them a client account, and leave them to it. The output would reflect that.
AI needs the same investment:
- A project or system with clear, specific instructions and full client context
- Access to real data — your keyword data, your GSC performance, your site’s content history
- Human review at each output stage, especially early on
- Updates as the client’s situation changes — a memory file that doesn’t get refreshed is a liability
For business owners reading this who are hoping AI will let them skip the SEO spend: the honest answer is that if you’re not prepared to invest time into setting this up properly, the risk outweighs the saving. The time and cost to fix AI-gone-wrong tends to equal — often exceed — what it would have cost to hire a competent person to do the work correctly.
The most overhyped idea in circulation right now is that a $20-a-month AI subscription replaces an SEO team. It doesn’t. What it can do is make an SEO team faster, more consistent, and better at handling volume. That’s a different — and genuinely valuable — proposition. But it requires the team to still exist.
What “ranking” means now
One more thing worth naming, because it changes the context of everything above.
Google AI Overviews now appear on 48% of queries, up from 34.5% six months ago. When they appear, organic click-through rates drop 38% for the pages that used to capture those clicks. Ranking #1 means something different than it did two years ago.
But here’s what hasn’t changed: SEO was never about gaming a single platform. It was always about optimising for how people find and evaluate information. Google, Bing, ChatGPT, Perplexity — these are all surfaces where people ask questions and look for trustworthy answers. The surfaces shift. The underlying principle doesn’t.
The sites that are holding up — and the ones appearing in AI-generated answers — are the ones that built genuine authority: specific, well-sourced, human-authored content with a clear point of view. That’s what AI can help you produce more of, more efficiently. It can’t substitute for it.
If you want to go deeper on optimising for AI search platforms specifically — getting cited by ChatGPT, appearing in Perplexity — the GEO layer is where to start.
The honest summary
AI is a real tool. Used correctly, it makes good SEO practitioners faster and more consistent. Used incorrectly — without context, without data, without human oversight — it’s an expensive mistake that looks like savings until it isn’t.
The question isn’t whether to use AI in your SEO work. It’s whether you’ve done the work to make it reliable. That means building the context layer. Wiring in the data. Keeping a human in the loop at every output that matters.
Most people skip those steps because they’re looking for a shortcut. The shortcut is the risk.
FAQ
Is AI-generated content bad for SEO?
AI-generated content isn’t automatically penalised — Google’s own guidance is that quality matters more than how content was produced. The problem is unedited, mass-produced AI output that adds no original value. Content where AI drafts and a human with real expertise edits, verifies, and adds specific insight performs fine. Content that skips that human layer is what gets hit.
What AI tools actually work for SEO?
The tools that work best are the ones connected to real data — Ahrefs, Google Search Console, your site’s crawl data. A standalone chat tool with no data access is guessing. The most effective setups I’ve built connect multiple data sources and hold client context in memory, so every task runs against what’s actually true about the site.
Can AI replace keyword research?
AI can handle a strong first pass at keyword discovery and clustering, especially with good context about the site and audience. It reduces sourcing time meaningfully. But the judgment layer — which keywords are actually worth pursuing given the site’s authority, the client’s commercial priorities, and the competitive landscape — still needs a human. AI surfaces candidates. An SEO decides what to do with them.
Why does AI keep making up URLs and internal links?
LLMs predict text based on patterns — they don’t check whether a URL exists. Even with a full list of live URLs provided, basic AI chat tools will fabricate plausible-looking links that don’t exist. For internal linking, always verify every URL against the live site before publishing. This is one of the most consistent failure modes I’ve seen in AI-assisted SEO work.
How should a business owner use AI for SEO without an agency?
If you want to use AI tools for SEO without professional support, the minimum viable setup is a project with clear instructions, full site context, and real data connected or manually supplied. Plan to review every output before acting on it. Understand that AI cannot produce an SEO strategy — it can help execute one. If you’re not prepared to invest time in the setup and oversight, the risk of getting it wrong likely outweighs the cost of hiring someone.
Does using AI for SEO help with AI search visibility too?
Not directly — these are separate problems. Using AI tools in your SEO workflow makes your traditional search optimisation more efficient. Getting your content cited by ChatGPT or Perplexity is a different discipline (GEO) that involves structured content, authoritative sourcing, and topical depth. The two reinforce each other, but they’re not the same thing.
No comment yet, add your voice below!