"Can AI do my SEO?" Owners ask it when a vendor promises the first page of the search results by spring, or when someone has started producing pages by the dozen and nobody is sure what they say. The honest answer is that a tool can draft pages and descriptions, propose titles, group the questions customers type, check the site for faults and watch where the company appears, and a person still decides what the company claims, which pages exist, whether any statement is true, and who holds the logins. This article is for the owner of a company with fifty to five hundred people and a website that is supposed to bring in enquiries. Where money is handled it is not accounting or tax advice, and what a search engine allows is a question for its own published rules and your advisor.

I was COO and then President of a public company that closed more than forty acquisitions, and I ran the integration on most of them. In almost every one, the acquired company's website was a surprise: pages nobody remembered writing, a service area that had not been true for years, and a ranking that rested on one old article a founder had written by hand. I am a lawyer by training. Now I write fixed-fee AI plans for CEOs of private-equity-backed and owner-led mid-market companies through 013 Labs.

The short answer

Search is a reputation held by a machine. A search engine shows your pages when it believes they answer what someone typed and that the company behind them is real and useful. A tool is good at the mechanical half: finding the questions people type, drafting a plain page that answers one of them from material you already have, writing the title and description, checking that the site loads, and counting which pages bring visitors. What it cannot do is decide what the company is willing to say in public, whether a page should exist at all, or whether a claim is still true. A page on your site is a statement by the company, and a search engine remembers statements for a long time. So the useful question is not "can a tool do our SEO" but "which pages are worth writing, who checks what they claim, and who owns the accounts".

What a tool can often take over

Grouping the questions customers type. Given your products and a few real enquiries, a tool can list the questions people put into a search box, group them, and show which ones no page on your site answers. That list is the most useful thing a tool produces.

Drafting a page from your own material. A plain answer to one question, written from your own notes and the way your people explain it on the phone. The draft is only as true as the material it was given.

Titles and descriptions. The line that appears in the search results and the sentence under it. A tool can propose them for every page; a person reads each one, because that line is the first thing a stranger sees.

Checking the site for faults. Broken links, pages that load slowly on a phone, two pages that say the same thing, images with no text for a screen reader. A tool can list them in minutes.

Watching where the company appears. Which questions bring visitors, which pages they land on, which pages nobody reaches. Can AI generate leads for my business covers what happens after the visit, and Can AI help promote my business covers where search sits in the rest of the marketing.

The neighbour: paid search. The same line holds for the advertising beside the results: a tool can draft the advertisement and suggest the words to bid on; a person sets the budget, approves every claim and decides each month whether the money is working. Not accounting or tax advice.

What a person must keep

What the company claims. The largest, the fastest, the only certified, a result for a customer, a guarantee. A page that promises an outcome the company cannot show is a liability everywhere. A claim is checked by the person who owns the fact before it is published, and again when the page is next touched.

Which pages exist. A tool asked for more pages will produce more pages. Fifty thin pages that say almost the same thing, one for every town in the county, are a risk, not a shortcut: they read as padding and drag the good pages down. A person decides a page is worth a stranger's time before anyone drafts it.

Whether a page is true for the place it names. Service areas, opening hours, which branch stocks what. A tool will happily write "same-day delivery in" followed by a town the company does not serve. Only a person knows the map.

The logins. The analytics, the search console, the host and the domain belong to the company, with two named administrators and a recovery address that does not leave when a person does. No tool or outside firm works from a free personal account or holds the only login to anything.

Links and reviews. Nobody buys links, and nobody invents a review, a quote or a customer; both are the oldest way to lose a search reputation. Can AI respond to my Google reviews covers the honest version.

Money and the bad day. An advertising budget, an outside firm's fee, a paid listing: decided by a person, measured against enquiries that turned into work, and stopped when they do not. Not accounting or tax advice. When the numbers fall or a page is found to be wrong, a person decides what comes down, and nothing new is published until that is done.

The data rule

A tool that works on your site asks for the keys: the host, the analytics, the search console, and often your customer enquiries. Write one sentence before any tool gets that access: which tool is approved, under which terms, what it may see, what it may publish without a person, and what never goes into it at all: a customer's name, a price list that is not public, anything confidential, anything about a deal that has not closed. What happens to your data explains what to ask a provider before a single file is uploaded, and Do I need an AI policy for my small business shows how short the written rule can be.

Five checks before a tool touches the site

First, write the claims file and the never-list. One page of every statement the company is willing to make in public, with the person who owns each fact, and a short list of what no page may contain: a guarantee, a price without a date, a town the company does not serve, an unchecked certification.

Second, decide which pages are worth writing. From the tool's list of questions, a person picks the ones a real customer would thank you for answering, and says no to the rest. One page per question, written to be read, is the whole method; our own site is built that way, one plain owner question per page, read by a person before it goes live.

Third, put the accounts in the company's name. The domain, the host, the analytics and the search console, each with two administrators and a company recovery address. Do this before any tool or outside firm is connected; whatever access they are given, they keep until someone takes it back.

Fourth, run draft-only for the first month. Every page, title and description is drafted by the tool and read by a person, free to change any word and to refuse the page. At the end of the month, count how many drafts went out unchanged and how many were stopped. Those two counts tell you how far to trust the tool.

Fifth, write the stop rule. Who can pull a page within the hour, who is told when the numbers fall, and who decides when publishing resumes. The person who may press that button has to be named in advance.

Check yourself

Seven yes or no questions. Each no is a thing to fix before a tool touches the site.

Mostly no? That is normal, and it is the real work. The AI readiness scorecard asks ten questions like these across the whole company.

Where it goes wrong

The pages by the dozen. A page for every town and every product, none written for a reader. Visitors left in seconds, the search engine noticed, and the good pages sank with the padding.

The claim nobody checked. "The region's only factory-certified service centre", lifted from an old brochure after the certification had lapsed. The competitor who now holds it noticed first.

The town the company does not serve. A page promised same-day delivery two hours from the nearest branch. The enquiries came, the promise could not be kept, and the reviews followed.

The invented quote. Asked for "a customer story", a tool produced one, signed with a first name and a town. No such customer exists. What is hallucination explains why a tool does this.

Starting safely

Start with the list of questions and the faults report, where nothing is published and the company only learns what people ask and what is broken. Then draft the pages a person has said yes to, one at a time. Then let the tool propose titles and descriptions for the pages that exist, with a person reading each one. Keep the claims file and the never-list; what changes is how much arrives already drafted. Can AI make a website for my business covers the site itself, Can AI do my social media the other public channel, Picking your first AI project the general order, and How to train employees to use AI the person who will run the tool.

Picture it going wrong

A made-up example, to show the shape. A family-owned industrial supply distributor, about two hundred people, three branches. The owner's son reads that a tool can "do the SEO", connects it to the website and the search console under his own login, and asks for pages for every product line in every town within a day's drive. By the end of the month there are several hundred new pages, and for a few weeks the numbers climb. Then an enquiry arrives from a town the company does not serve, asking about the same-day delivery the page promised. Then a manufacturer asks why the site calls the company its only authorised distributor in a state where it has three. Then the numbers fall, hard: the search engine has decided the site is mostly padding. When the owner asks for the pages to come down, nobody can say which carry a claim, because nobody read them on the way out, and the search console sits under the son's login. Nothing was wrong with the tool. Nobody had written what the company claims or approved anything, and the accounts were in the wrong name. The fix was the accounts in the company's name, a claims file with an owner for each fact, every unapproved page removed, real customer questions answered one page at a time and read before publishing, and a stop rule with a name on it.

Mistakes that repeat

Frequently asked questions

Can AI do my SEO?

A tool can group the questions customers type, draft a page that answers one of them from your own material, propose titles and descriptions, check the site for faults and watch where the company appears. A person decides what the company claims and which pages exist, and holds the logins.

Can AI do SEO for my website on its own?

Some tools will publish without anyone looking. That is the setting to turn off. Draft-only mode, with a named approver, gives you the steady work without the public mistake.

Can I use AI-written content for SEO?

Yes, when it is drafted from material you gave it, answers a question a real customer asks, and is read by a person before it goes live. The failure is volume for its own sake: thin pages produced in bulk are a risk to the whole site, not a shortcut.

Are FAQ pages good for SEO?

A page that answers the questions customers actually ask, in their words, is good for the reader, which is what a search engine tries to reward. A list of questions nobody asks is not.

Can AI do my Google ads as well?

It can draft the advertisement and suggest the words to bid on. A person sets the budget, approves every claim and decides each month whether the money brings enquiries that turn into work. Not accounting or tax advice.

Is it safe to give an SEO tool access to my website and analytics?

Only under written terms that say what the tool may see and what it may publish, through company accounts, never a free personal account. Private price lists and anything confidential stay out of the tool altogether.

Where should I start?

Put the accounts in the company's name, write the claims file and the never-list, choose the questions worth answering, run the tool in draft-only mode for a month, and write the stop rule before the first page goes live.

Not sure where you stand? Take the 3-minute AI readiness scorecard.