AI is good at SEO ideas, drafts and wording, and bad at the numbers SEO depends on. AI Central's Do Your SEO With AI in 7 Steps splits the work along that line, from a keyword map to an old-post refresh, and every prompt is built so search volumes and facts come from your SEO tool or Search Console, never from the chat.
The invented number problem
The guide's opening example asks for 10 keywords for a physio clinic booking app with monthly search volume. The answer comes back with exact figures, such as 12,400 a month for clinic booking software, that nobody checked and that the AI cannot know. The second failure is quieter: ask for an article and you get the same page as everyone else.
Numbers nobody checked
Steps 1 and 2: from buyer to page plan
Step one maps keywords by intent, not by volume. You give the product, the buyer and their problem, and ask for 30 searches grouped into learn, compare and buy. The prompt says it outright: do not guess search volumes, I will paste them from my keyword tool. Then you search the top ideas yourself.
Step two groups those searches into topics, one page per question. Paste the searches with their real monthly volumes and ask for clusters one page can answer, each with a main keyword, intent and working title. Paste your existing URLs too, so the AI can flag any cluster a page already covers, which is how you avoid two pages chasing the same search.
Steps 3 and 4: brief first, draft second
The brief is where the guide puts the effort. The prompt names the keyword and the reader, then asks the AI to search the web, read the top 5 results, list what they all cover and what they miss, and only then write a brief with an angle, sections and 5 reader questions. You add your own data at the end, in the example a line about SMS reminders, because your own data is what fills the gap the top results leave.
The brief is where you beat the copies
The draft follows the brief one section at a time. The prompt asks for an outline with H2 and H3 headings, a note under each H2 on which proof to add from your own clients, and a first section of 150 words in plain language. Every claim you must check gets the marker [CHECK]. The guide backs this with Google's own advice: fact-check AI content before you publish.
Steps 5 to 7: fix the pages you already have
Titles and descriptions are written for the person who clicks. Main words go first because long titles get cut off, each title is unique on your site, each keyword appears once, and there is no clickbait. The prompt asks for 3 options of each and for the pair the AI would pick and why. Google can rewrite them, so each one must match what the page says.
Internal links help readers and Google reach your pages. Paste the new post and a list of your URLs with titles, and ask for up to 5 links, each with the sentence, the anchor text, the URL and why a reader would click. Anchors must describe the page, never click here.
The refresh starts from Search Console, not from the calendar. Export the page's queries for the last 3 months, find the ones with many impressions and few clicks, and ask which of them the post answers badly. The AI then suggests new sections and facts to update while you keep what still works. Changing only the date does not count as an update.
Where each piece of data comes from
- Keyword ideas: ask the AI
- Search volumes: your SEO tool
- What ranks today: web search, or look yourself
- Clicks and impressions: Search Console
Ideas from AI, numbers from real data
What to watch out for
- Never accept a search volume the chat produced on its own. Paste the real one.
- Search the top ideas yourself before you commit a page to them.
- Resolve every [CHECK] marker before you publish.
- A refresh changes facts and sections, not only the date.
Where to start
The closing card names the tools, ChatGPT, Claude, Gemini or Perplexity, and two first jobs: map one topic cluster and refresh one old post. Both jobs start from data you already have: your keyword tool export for the cluster and Search Console for the old post.