Before Bolting AI Onto Your SEO Stack, Answer This One Question First

A while back, I had dinner with a friend who runs growth.

He pointed at a table full of food and complained to me: "Liu Run, last year our company bought four AI tools — one for keywords, one for writing, one for technical audits. And you know what? Traffic didn't budge. We did crank out a pile of content, but nobody's even reviewing it."

I didn't say anything after hearing that.

Because this is the third time this year I've heard this kind of story.

90% of teams are using AI — but "using" and "using it right" are two different things

There's a data point that's pretty interesting: over 90% of marketing teams now have AI somewhere in their workflow.

Hearing that number, your first reaction might be: wow, that high already? I need to get on this fast.

But wait.

"Having AI in your toolbar" and "having the right AI in your toolbar" are worlds apart.

What does "the right AI" mean?

Let me break it down in one sentence: pick the tool based on the job you need done, not on which tool is the hottest.

That sounds like a platitude, but in practice, 90% of teams do it backwards. They look at what everyone else is using, then buy the same thing. Only after buying do they realize: this tool doesn't even connect to our workflow.

What can AI actually do in SEO?

To pick right, you first need to know what AI can actually help with in the SEO game.

I tallied it up — it mostly comes down to seven categories of work:

1. Keyword research and clustering. It used to be one keyword at a time into a spreadsheet. Now it's auto-grouping related terms by topic and intent.
2. Content briefs and drafting. Based on top-ranking pages and search intent, it generates structured outlines — even entire first drafts.
3. On-page optimization. Scores your existing content and tells you where you're missing terms, where the structure is off, where it's hard to read.
4. Technical audits. Crawls your site and surfaces every technical issue — dead links, slow loading, duplicate content.
5. Internal linking. Across hundreds or thousands of pieces of content, it finds linking opportunities you should have but don't.
6. AI search visibility. This one's new. It monitors whether your brand shows up in the answers from AI engines like ChatGPT, Perplexity, and Google AI Overviews.
7. Reporting. Mashes data from a pile of channels together and tells you what's actually driving organic traffic.

Of these seven, the highest ROI sits with the work that's repetitive, high-volume, and rule-based.

Things like keyword clustering, technical audits, on-page optimization — these need doing anyway. Doing them by hand is slow and prone to gaps. Get AI on it and the work starts flowing immediately.

On the flip side, things like content drafting and strategy that need judgment — AI can help, but it can't be the main player. More on that later.

Six questions to answer before you reach for your wallet

Alright. Now you know what AI can do. But what should your team actually buy?

Here are six questions. Run through every one of them before buying any tool.

First: which problem are you most urgently trying to solve?

Is it building topical authority? Making your site easier for crawlers to crawl? Publishing 50 more pieces a month?

Those three goals map to completely different tools. Buying a tool before you've figured out what you're trying to do is like walking into a supermarket without knowing what you want to eat. You come out with a full cart, but nothing in it is what you actually need.

Second: how deep a technical diet can your team stomach?

Some tools are built for specialists. A crawler like Screaming Frog is technically very deep, but it assumes you can read crawler data. If no one on your team can interpret this stuff, buying it is just decoration.

Other tools go the opposite way — they're built for "people doing marketing who don't understand SEO" and just hand you a "do this" recommendation without explaining why.

Third: how many pieces do you publish a month?

If you publish five pieces a month, standing up a heavy automation stack is using a sledgehammer to crack a nut. The maintenance cost will exceed the output.

If you publish 50 or more a month, automation is worth it. Briefs, clustering, on-page optimization — once these scale up, the compounding returns kick in fast.

Fourth: does the tool connect to your CMS and CRM?

A disconnected tool means disconnected data.

A tool that plugs straight into your CMS means its suggestions can be acted on in the moment, while you're writing the content — not by jumping to another dashboard, looking at it, then coming back to check item by item.

A tool that connects to CRM goes even further. It can tie SEO results directly to your pipeline and revenue, instead of just handing you a traffic number.

Fifth: who reviews your content?

AI-written first drafts and auto-generated optimization suggestions all need a human gate. Especially on things that can't go wrong — brand voice, factual accuracy, compliance.

When picking a tool, pick the kind that's designed with "human review" built into the workflow. The kind that defaults to "auto = publish directly" will get you into trouble sooner or later.

Sixth: does the tool's reporting show vanity metrics, or real business?

Traffic that looks nice but doesn't convert adds up to zero.

A reporting layer that's actually worth something tells you: did all this SEO work turn into qualified traffic, into leads, into money?

Don't expect one tool to do it all

I have to bold this one:

No single AI SEO tool does all seven categories of work well.

The smartest move is to assemble a stack by workflow stage. Put the best-in-class for each stage.

Let me walk you through it by task. You don't need to memorize it — just remember the thinking.

The research and clustering stage

Doing keyword research used to mean one person staring at a spreadsheet, typing in keywords one at a time. Tiring, slow, and even when you finished you had no idea how those words related to each other.

Today's AI tools can do two things: pull all related terms out, and group them automatically by topic and search intent, even sorting them by difficulty. Done in minutes.

The capability you want at this stage is clustering. Which terms can be merged into a single page, and which need their own separate piece. That's the judgment that actually saves time.

Some tools cluster based on how Google itself treats synonyms on the search results page (SERP). Not just topic-based grouping. That's clustering that's actually useful — not just lumping words together by surface similarity.

Audience research: an underestimated dimension

Most SEO tools start from "keywords" and work backwards to "who's searching."

But one category of tools does it the other way. It starts from "your target audience" and tells you what podcasts they listen to, what sites they read, what YouTube channels they watch, what social accounts they follow.

Keywords tell you "what they're searching for." Audience research tells you "where their attention lives."

These two map to completely different channels. When it comes to content distribution and PR, the second one is far more useful than the first.

Question trees: mapping out "what else do people ask"

One category of tool pulls Google's "People Also Ask" data directly and visualizes how one question branches out into the next, as a tree.

Why is this useful?

Because that tree reflects how Google itself understands the relationships between topics. Not guessed by AI — grown out of real search behavior.

For building FAQs, designing topic clusters, finding content gaps, this beats a keyword list by a mile.

The content brief and drafting stage

A content brief, plainly put, is a list you hand writers — "here's what to write about, which points to cover, who to reference" — so they don't start from zero.

AI can accelerate this stage: based on top-ranking content and search intent, it generates structured outlines, even entire first drafts.

But there's a huge trap here, and I have to spell it out. More on this below.

The on-page optimization and content refresh stage

This is the stage where AI tools matured earliest and where ROI is most direct.

It takes your content, compares it against top-ranking competitors, and tells you: which concepts you should have covered but didn't, where the structure is loose, how to adjust keyword density.

Some tools score by "topical relevance" rather than just stuffing keywords. Those are more reliable than tools that just force-fit words.

Other tools go even further. They scan your entire content library and tell you which piece needs a refresh, which two pieces are cannibalizing each other's traffic, which topic you haven't touched at all. This is for strategic planning of large content libraries — not patching up a single article.

The technical audit stage

Technical SEO — crawl efficiency, site speed, structured data, dead links, redirects — has to be handled by tools that can read your site the way a search engine does.

There are two approaches here.

One is for technical SEO specialists: crawl once, give you all the data, extremely deep, but assumes you can read it.

The other is for people who "need to report results to a non-technical boss": translates crawler data into plain English, sorts it by impact, generates a report you can take straight into a meeting. The audiences for these two are completely different.

Whichever one your team is, buy that kind of tool. Don't buy the wrong one.

The internal linking stage

Internal linking — high ROI, low cost — but 90% of teams underdo it.

Why? Because manually finding the links you should have across hundreds or thousands of pieces of content is basically impossible.

AI tools can do two things here: discover "should-link-but-doesn't" opportunities in your content library, and some even work inside your editor — reminding you in real time as you write that "you could link this part to that piece."

Internal linking is the kind of work where doing it pays off immediately, and not doing it means you're constantly bleeding traffic. If a tool can automate discovery, don't muscle through it by hand.

AI search visibility: a brand-new battlefield

This stage only emerged in the last couple of years.

Traditional SEO watched Google rankings. But now more and more people ask ChatGPT, ask Perplexity, or check Google's AI Overviews before buying something. How these AI engines talk about your brand directly decides whether you get seen.

This is called AEO — Answer Engine Optimization.

AEO and SEO are two disciplines, but complementary. Teams that do both at once are the ones who hold their ground in the channels where buyers look for answers.

The interesting part is that some tools can already monitor, across the various AI engines' answers, on which topics your brand gets cited — and on which topics you don't show up at all. That's a layer traditional SEO reporting doesn't cover.

Don't let automation tank your content

Alright, tools covered.

But tools are only amplifiers. Amplifiers amplify what you already have. If what you have is garbage, the amplified version is still garbage — just a much larger pile of garbage.

Done badly, automation amplifies three things at once: thin content, wrong intent, scrambled tone.

So how do you avoid face-planting?

Here's a six-step process.

Step 1: Research in bulk, don't chip away at one keyword at a time

Whatever tool you use, start by building one big keyword list. Then drop it into a clustering tool and group by topic and intent.

The point of this step is to see the full landscape of your topics first, then decide what to write.

The biggest fear is starting to write without clustering first — then writing three pieces and realizing they're all cannibalizing each other's rankings. This is called keyword cannibalization, the most common trap for beginners.

Step 2: Map every cluster to a content type

Once clustering is done, map each group to one of three content types:

• Pillar content: broad, authoritative, targets big terms
• Supporting content: specific, long-tail, answers small questions
• Landing pages: built for conversion

This mapping is the skeleton of your editorial calendar. Not a keyword list — a content architecture.

Step 3: AI drafts are a starting point. Never publish directly.

This is the most important step.

An AI-generated first draft is a structured research document — not a final draft.

Treat it as a starting point. Then you absolutely must do these things:

• Fact-check
• Align with brand voice
• Add your own examples, your own cases, your own insight

These are the things AI can't give you.

Publishing an AI first draft straight is the dumbest way to use it. Yes, you have content. But you haven't built authority. Neither search engines nor readers will buy it. What actually makes content worth citing is that layer of human craft on top.

Step 4: Reviews must have a checklist and a named owner

Every workflow that uses AI drafting must have a human-review checkpoint. This is not optional.

Write it into your SOP. Make clear:

• What the reviewer is checking for
• What standard a draft has to meet to pass
• Who owns that decision

At scale, use a checklist. Build it once, use it every time.

Step 5: Put audits on the calendar the way you put production on the calendar

"Publish and forget" is the biggest traffic leak there is.

Run a content audit once a quarter or every half-year. Which pages have dropped in rank, which data has gone stale, which two pieces are cannibalizing each other. The audit tool will tell you.

Schedule audits as seriously as you schedule production drafts. Don't wait for traffic to fall before you remember to check.

Step 6: With every audit, do an internal-link pass while you're at it

Every content audit should include an internal-link check.

• Which pages have zero internal links pointing to them? (Orphan pages)
• Which pillar pages should link to core supporting content but don't?
• Which semantically related linking opportunities are still missing from your library?

Internal linking is one of the highest-ROI, lowest-cost pieces of SEO work — and one of the most neglected. Doing a pass with every audit is ten times easier than scrambling to fix it after the fact.

A few questions I get asked a lot

Can AI-written content rank?

Yes. But only if it has original insight, accurate information, reasonable structure, and genuine subject-matter expertise in it.

The problem has never been "AI wrote the first draft." The problem is AI first drafts being published without meaningful human review.

What search engines look at is whether the final piece is useful and demonstrates expertise — not how it was produced.

How often should I refresh content?

Commercial terms, comparison content, anything involving data and tools — audit at least every six months. Slower-moving evergreen educational content, once a year is fine.

In practice, don't be a slave to the calendar. Use audit tools to flag pages whose rank has dropped or whose content has gone seriously stale, and queue those for refresh based on that signal. It's far more efficient than a blanket schedule.

How should a small company choose?

Small companies shouldn't bite off more than they can chew. Cover three key stages: research, on-page optimization, technical audit. One usable tool per stage beats a pile of ten you can't actually use.

Prioritize tools that someone with no specialized SEO background can pick up. Don't buy a bunch of features you won't use just to look professional.

Tools are amplifiers. Human judgment is what gets rewarded.

After all these tools, I want to close with something that might feel a little counterintuitive.

AI SEO tools are table stakes today. Don't buy them and you'll fall behind. But the tools themselves won't make you win.

What makes you win is what you do with the time the tools save you.

Do human judgment. Original insight. Experience no one else has.

That's what search engines and AI systems actually reward.

Tools can compress your research from 17 hours down to 2. But if you take those 15 saved hours and pump out more AI first drafts, you'll lose badly.

Use them to think through an angle no one else has thought of. To build a case study no one else can build. To spend half an hour talking to a real user.

That's where those 15 hours should go.

Tools make you faster.

But faster was never the point.

Figuring out which direction to be fast in — that's the point.