Most of what gets sold as SEO content now is a prompt and a publish button. Somebody feeds a keyword to a model, takes the first thing that comes out, and puts it on your site under your name. It reads like nobody has ever done the job it describes, because nobody has, and both Google and your customers have gotten good at spotting it.

I use the same tools. The difference is everything around them. I do the search research, I get you on the phone and take down what you already tell customers every day, I draft with AI doing a share of the typing, and then I go through every line and cut what is wrong, vague or padded. Nothing goes live that I have not read end to end. If you read a sentence on your site and want to know why it is there, you can ask me and get a real answer.

This page was written that way. So was every other page on this site. I would rather tell you that up front than have you wonder about it.

Volume matters, but only after quality does

A site that adds useful pages on a regular schedule looks alive to Google. A site that has not been touched in three years looks abandoned, and it gets treated that way. That much is not controversial.

What gets people in trouble is treating the schedule as the goal. I have taken over sites that were publishing four articles a month for two years straight, all of it thin, all of it written to a keyword list by somebody who had never seen the business. Six hundred pages of nothing. Google's spam policies have a name for that now, scaled content abuse, and the March 2024 core update took a lot of those sites apart.

So the honest answer on volume is: enough to stay current in your niche, never more than you can do well. For most Tucson service businesses that means one strong page per service, one per city you genuinely work in, and then a steady trickle of articles that answer the questions you already field on the phone. If I cannot tell you who a page is for and what question it answers, I do not write it.

Search intent decides what the page says

Before I write a word, I work out what the person doing the search actually wanted. There are four flavors of it and they need four different pages.

• Informational. They want to understand something. "why is my ac short cycling"
• Navigational. They want a specific site or business. "promenade dental murrieta"
• Commercial. They are comparing before they buy. "best junk removal temecula"
• Transactional. They are ready. "junk removal near me"

Get this wrong and no amount of writing saves you. A sales page will not rank for a question, and an article will not rank for a "near me" search. The fastest way to check is to search the term yourself and look at what Google is already rewarding. If the first page is all how-to articles, Google has told you what kind of page it wants. Arguing with that is expensive.

Keywords come after that decision, not before it. They go in the title, the H1, one or two subheads and the opening paragraph, and then I stop thinking about them and write like a person. Google has understood synonyms and context for years. It has not needed you to repeat a phrase eleven times since about 2011.

What already ranks tells you what Google wants

The top ten results for your term are not your enemy. They are a free brief. Google has already tested those pages against millions of searches and kept them, so whatever they have in common is what it currently believes the answer looks like.

I read all of them. Not skim, read. I write down every subheading, every question they answer, every thing they cover that you do not, and, more usefully, every question they all skip. That last list is where the openings are. The gap is almost always something local, something about price, or something a person who has done the work would know and a writer working from a keyword list would not.

Then I write a better page. Not a longer one. A better one.

Original content, taken out of your head

Google's raters are told to look for evidence that content was produced by someone with real experience of the subject. That is the first E in E-E-A-T, and it is the part that cannot be faked by research.

So I do not research your industry and write at you. I interview you, usually 30 to 60 minutes on the phone, and I record it. What comes out of that call is the stuff nobody else can publish: the objection you hear every week, the reason the cheap option fails in July, the mistake customers make before they call you, the price band and what moves it. I have written for junk removal and for dentistry in the same month, and in both cases the pages that ranked were the ones built almost entirely out of things the owner said on that call.

If a claim cannot be traced back to you or to a document I can point at, it does not go on the page. That rule costs me some nice-sounding sentences. It also means nothing on your site can be knocked down by a customer who knows better, and it is the rule that makes using AI safe rather than reckless.

How I use AI, and what it is not allowed to do

Most agencies use these tools and will not say so. I would rather be direct, because if you are paying for content you are entitled to know how it gets made.

Google's own position is that content is judged on quality, not on how it was produced, and it is right about that. A page is good or bad on its merits. The failure mode has never been the tool. It is the absence of anyone doing the thinking.

Where it helps

• Reading and summarizing the top-ranking pages for a term faster than I can get through them by hand.
• Turning a recorded interview with you into a first draft that already has your facts in it.
• Drafting sections once I have decided what the page needs to say and in what order.
• Checking my own work: catching repetition, testing whether an answer actually answers the question, confirming the FAQ text on the page matches the FAQ text in the schema word for word.
• Grinding through the parts where fatigue causes mistakes, such as alt text, structured data and internal link audits.

What it is not allowed to do

• Supply facts about your business. Those come from you, on a recorded call. A model does not know your pricing, your service area, or why customers leave your competitor.
• Invent a statistic, a study or a source. If a number appears on your page, I can show you where it came from.
• Write a testimonial, a review, or anything in a customer's voice. Not once, not ever.
• Decide search intent or page strategy. That is judgment, and it is the part you are actually hiring.
• Publish unread. There is no step in my process where text goes from a model to your site without passing through me first.

The line Google enforces is scaled content abuse: pages mass-produced with little human involvement in order to game rankings. The distance between that and what I do is not which software is open. It is that a person decided each page should exist, sourced what is in it, cut it down, and is accountable for every line of it.

Keywords and the on-page work around them

Writing is half of it. The other half is everything wrapped around the writing, and it is where most content projects quietly fail because the writer hands over a Word document and somebody pastes it in.

On every page I write, I also handle the title tag and meta description, the heading hierarchy, the internal links in and out, image file names and alt text, and the structured data that tells Google what the page is. On this site the FAQ text you can see and the FAQ text in the schema are matched word for word, because a mismatch is how you get a manual action instead of a rich result. That is not an extra service. It is what writing a web page means.

If the underlying site has problems that no amount of good writing will fix, that is a different job, and it is called technical SEO.

Structure: written to be read, and to be quoted

Nobody reads a web page. They scan it, decide in about eight seconds whether it is going to answer them, and either commit or leave. Nielsen Norman Group has been publishing research on legibility and comprehension for two decades and the finding has never changed.

So the pages I write are built to be scanned. Descriptive subheads instead of clever ones. Paragraphs of two to four sentences. Lists where the content is genuinely a list and not where it is prose I could not be bothered to write. The answer to the question in the heading appears in the first sentence under the heading, not in the fourth paragraph after a warm-up.

How your content gets cited by ChatGPT, Claude and Google's AI answers

A growing share of the people looking for what you sell will never see a list of blue links. They will ask an assistant and read one paragraph. That paragraph is assembled from pages, and there are things you can do to make yours one of them.

What I have found works, and what I now build into every page:

• Answer in the first sentence. A model extracting an answer takes the sentence directly under the heading. If yours is a wind-up, it takes somebody else's.
• Make each section self-contained. An assistant may lift one FAQ answer with no surrounding context. If that answer only makes sense after reading the three paragraphs above it, it is useless when quoted.
• Name things explicitly. Write "Tucson SEO, run by David Cragg in Tucson, Arizona" rather than "we" and "here." Pronouns are invisible to a system trying to work out who the page is about.
• Be specific and be checkable. Real figures, real constraints, real dates. Vague marketing copy gets skipped in favor of a page that commits to something.
• Match your schema to your visible text. Structured data is how a machine confirms it read the page correctly.

None of this is a trick, and none of it is separate from writing well. It is the same discipline: say the true thing, early, in plain words. There is some irony in using these tools to write pages that these same tools go on to cite. I will take it.

How I tell whether a page is working

Most content reporting is theater. Somebody sends you a chart of pageviews and calls it a result. Here is what I actually watch, in order.

• Impressions and average position in Google Search Console. This moves first, usually well before traffic does. A page that has gone from position 34 to position 14 has not sent you a single visitor yet, and it is working.
• Which queries the page is picking up. If it is collecting impressions for things you do not sell, the page is answering the wrong question. That needs a rewrite, not more links.
• Clicks and click-through rate. Impressions with no clicks is almost always a title tag problem, which is a twenty-minute fix rather than a content problem.
• Whether the phone rings. In the end this is the only number that pays for anything, and it is the one I ask you about.

What I put very little weight on: bounce rate, time on page and raw pageviews. Bounce rate gets quoted constantly and means almost nothing on a service page, where a reader who got their answer and picked up the phone looks identical to one who left in disgust. A metric that cannot tell those two apart cannot tell you whether the writing is any good.

What I will not do

Short list, and I will hold to it.

I also will not keyword stuff, spin an existing article, generate pages in bulk to hit a quota, publish anything under your name that you have not read, or pad a page to reach a word count. If a page does not need to exist, the right move is to not write it, and I will tell you that even though it is the version of the answer that bills less.

Where content fits with the rest of the work

Content on its own is half a strategy. It needs somewhere sound to live and something pointing at it.

Want to see how it plays out over months rather than in theory? The case studies walk through real client sites, including the air conditioning company whose reach I grew across the Tucson metro. The blog is where I write about what is working right now.

Who actually writes it

Written by David Cragg, SEO consultant, Tucson, Arizona. In search since 1990. Reviewed and updated August 2026.