How to Write Website Articles With AI That Still Rank on Google

AI-assisted articles can rank, because Google judges the page rather than the tool that wrote it. Use AI to research, outline and draft, but build each article on evidence only you have: your own numbers, photos, customer questions and first-hand checks. Publish a few strong pages, not dozens of near-identical ones, and have a named person verify every fact.

The mistake that sinks most AI articles happens before the prompt. If the only input is a topic, the model returns the average of what is already on the web, and Google has no reason to rank a tenth copy of page one. The fix sits in the material you feed in, which is why most of this method happens before and after the AI step.

Follow me on Instagram@sagnikteaches

What Google says about AI-written articles, in plain terms

Google's guidance on generative AI content is short and worth reading once. It says AI can be particularly useful for researching a topic and adding structure to original content. Its warning is about volume without value: using AI to generate many pages that add nothing for readers can breach the spam policy on scaled content abuse, which also covers stitching together other sites' content and "synonymising" existing pages.

Connect on LinkedInSagnik Bhattacharya

Three other points from Google's own pages shape the method below:

Subscribe on YouTube@codingliquids

• No preferred word count. Google's helpful-content page asks "Are you writing to a particular word count…?" and answers "(No, we don't.)" Length is a result of answering fully, never a target.
• Who, how and why. Google suggests clear bylines and author background, and explaining how automation was used where readers would reasonably wonder.
• No special trick for AI Overviews. Google says its AI search features need no optimisation beyond normal SEO, and that llms.txt files and special markup aren't used for them. A page that ranks well is the page that gets cited.

The phrase to hold on to is "original information, reporting, research, or analysis", taken from Google's self-assessment questions. An AI model cannot supply that. You can.

Choose questions from your own inbox, not from a keyword tool alone

Picture a holiday-let manager looking after 22 cottages for their owners. Their website gets steady traffic to the property pages and almost none to its generic "things to do" pages. The best article ideas are already sitting in their inbox and booking notes, because guests ask the same questions before they book.

Spend 30 minutes listing the last three months of pre-booking questions, then check each against the Queries tab in Google Search Console (Performance report) to see whether people already find you for related searches. Keyword research tools help size demand; keyword research with AI for a small business website covers that side. But the filter that matters is the last column below.

Anything where the third column says "nothing" is a page AI would write from the same sources as everyone else. Leave it.

Build an evidence pack before you open the AI tool

An evidence pack is a single document of facts the article must use and must not contradict. It takes 30 to 60 minutes and it is the difference between a ranking page and filler. Here is a filled-in version for the first article on the list:

EVIDENCE PACK: "Dog-friendly holiday cottages: 11 checks before you book"
Author: [manager's name], manages 22 cottages, walks every garden at handover
Hard facts (from our records, checked 12 Sep):
- 9 of our 22 cottages accept dogs; 3 limit dogs to the ground floor
- Enclosed gardens: 5 fully enclosed, 2 fenced on three sides, 2 no garden
- Dog fee: $25 per dog per stay at 7 cottages; free at 2
- Maximum dogs: 2 at most properties, 3 at one farmhouse
Guest questions (inbox, last 3 months): "Can dogs go upstairs?" (14),
  "Is the garden secure?" (31), "Do you provide bowls/towels?" (9)
Real problems we've seen: gate latch failing at one cottage (fixed Aug);
  guests assuming "dog-friendly" meant dogs allowed on sofas
Photos available: garden boundary shots for all 9, boot room, dog welcome kit
Must NOT claim: that any beach or footpath allows dogs (rules vary; link out)
Our opinion: ask for a photo of the boundary, not the lawn

Notice the last two lines. The "must not claim" list stops the AI adding confident general statements you can't stand behind, and the opinion line gives the article a point of view, which generic drafts never have.

Prompting for an outline, then a draft you can edit

Ask for the outline first. It takes two minutes to fix an outline and twenty to fix a draft built on a bad one.

You are helping [business] write a website article for guests.
Question the article must answer: [exact question]
Reader: [who they are, what they're about to decide]
Use ONLY the facts in the evidence pack below. If a point needs a fact
that isn't in the pack, write [NEED FACT: ...] instead of guessing.
Step 1: propose an outline of H2 headings, each phrased as the reader
would ask it. Put the direct answer in the first paragraph.
Do not add an introduction section or a conclusion section.
[paste evidence pack]

An illustrative outline the model might return, with the edits I'd make:

1. What does "dog-friendly" actually mean at a holiday cottage?
2. Is the garden fully enclosed? (use our 5 / 2 / 2 split)
3. Where can dogs go inside the cottage?
4. What does a dog cost per stay?
5. What's provided: bowls, towels, beds
6. [NEED FACT: nearby vets and emergency numbers]
7. The 11-point checklist to send the owner before you book
8. The benefits of travelling with your dog      <- delete: filler
9. Conclusion                                   <- delete: not asked for

The [NEED FACT] marker is the most useful line in the prompt. Here it flagged something a guest would genuinely want and the manager hadn't thought to include. Add the fact, or cut the section; never let the model fill it in.

Then ask for the draft section by section, pasting your edited outline back in. Drafting in chunks keeps the model close to the evidence pack; a single 2,000-word request drifts into general advice by the halfway point.

The edit that turns a generic draft into a page worth ranking

This is where the hours go, and where the ranking is won. Compare the paragraph a model typically writes with the one the manager ended up publishing:

AI draft: Travelling with your furry friend can be a wonderful experience. Many holiday cottages welcome dogs, but always check the property's pet policy before booking. Look for features such as a secure garden and nearby walks.

Edited: Of the 22 cottages we manage, nine take dogs, and "dog-friendly" means something different at each. At three, dogs stay downstairs. At two, the garden is fenced on three sides and open to a field on the fourth. Before you book anywhere, ask for a photo of the garden boundary, not the lawn.

The edited version has numbers, a specific trap and an instruction. Nobody else can write it. Work through each section with four questions:

1. What did we see or measure here? Add it. Numbers from your records, a problem you fixed, a photo you took.
2. Would a competitor's page say the same sentence? Cut it or make it specific.
3. Is every claim traceable to the evidence pack or a source? Run the routine in fact-checking AI marketing copy on anything that isn't.
4. Does it sound like us? A written brand voice guide for your AI tools gets the draft closer, but read it aloud anyway.

Strip the phrases that mark text as machine-written: "furry friend", "whether you're a… or a…", "a must-visit", paragraphs that open with "When planning…", and closing lines that summarise what the reader just read. A shared list of phrases your business never uses saves arguing about these each time.

Finally, add the things only a real business has: original photos with descriptive alt text, a byline linking to a short author page, and the date you last checked the facts. The author box and a one-line process note can be as simple as this:

Written by [name], who manages our 22 cottages and walks every garden at each handover. Drafted with an AI assistant from our own property records and guest questions, then checked and edited by [name]. Facts last checked 12 September.

That covers Google's "who" and "how" suggestions in two sentences, and it tells a guest why this page is worth trusting over a directory listing.

Original analysis: the article only your records can produce

The second article on the list, "Is a cottage cheaper than a hotel for a family of five?", shows the other kind of first-hand value: a calculation nobody else can run with your numbers. Give the AI your real price data and ask it to do the arithmetic and lay out the comparison, then check the sums yourself.

Here are our real weekly prices for 3-bedroom cottages by season, and
typical hotel prices the family would face (two rooms needed).
Build a comparison for a family of five, 7 nights, showing:
accommodation, breakfast, evening meals (self-catered vs eating out),
parking. Show every calculation. Flag any figure you had to assume.

An illustrative result for a school-holiday week, after the manager corrected two assumptions:

The model first assumed one hotel room would do for five people, and priced evening meals from a national average it made up. Both were caught because the prompt told it to flag assumptions. The finished article leads with the honest answer ("usually yes for a week, often no for two nights, because of cleaning fees and minimum stays") and shows the table. That nuance, from someone who sees the bookings, is what makes the page more useful than the generic "cottages offer great value" pages it competes with.

Titles, descriptions and links: where AI drafts go wrong

Models write titles for clicks in general, not for the query in front of you. A typical suggestion and a better one:

• AI: "The Ultimate Guide to Dog-Friendly Holiday Cottages: Everything You Need to Know"
• Better: "Dog-Friendly Holiday Cottages: 11 Checks Before You Book"

The better title matches how people search, promises something concrete and is short enough to display in full. Write the meta description yourself from the first paragraph, keep it roughly 120 to 155 characters, and make sure it answers the question rather than teasing it. Google asks that titles, descriptions, structured data and alt text are accurate when automation is involved, so check the model hasn't promised a checklist the page doesn't contain.

Internal links are where small sites lose most. Link the article to the property pages it mentions, to the booking page, and from two or three older pages back to the new one. AI will happily invent link targets, so paste your real page list into the prompt if you ask it to suggest links.

Why forty area guides in a week backfires

Here is the failure I'd expect if the same manager took the shortcut. They ask a model for a guide to every village near their cottages, forty pages in an afternoon, each built from the same prompt with the place name swapped. Each page reads fine on its own. Together they share a structure, the same eight attractions described in the same words, and no first-hand detail.

What typically follows: most of the pages sit in Search Console's Page indexing report under "Crawled – currently not indexed" or "Discovered – currently not indexed", meaning Google found them and chose not to include them. The few that are indexed get impressions at position 40 and beyond. And the site now carries forty weak pages that make the whole domain look thinner, which is the pattern the scaled content abuse policy exists to catch.

The better version of the same idea is five guides for the villages where you actually manage cottages, each with a walk you've done, a pub you'd send an owner to, and photos from the property. If you already have pages like the forty, an audit of website content that needs fixing helps you decide which to merge, improve or delete.

Reading Search Console at four, eight and twelve weeks

New articles on small sites can take weeks to settle, so judge them on a schedule rather than daily. Open the Performance report, filter by the page's URL, and compare against this table:

For the dog-friendly article, an encouraging week-eight picture would be impressions for a dozen variations of "dog-friendly cottage enclosed garden" and a handful of clicks to the matching property pages. That tells you the evidence-led format works and is worth repeating for the next question on your list. If AI assistants are part of how guests find you, an afternoon SEO audit with AI is a good quarterly companion to this check.

Time and cost for a realistic publishing rhythm

For one person writing, a sustainable pace is two to four evidence-led articles a month. Each takes roughly five hours: one hour on the evidence pack, 30 minutes of prompting, two to three hours of editing, and an hour on photos, links and publishing. The tool cost is a single chat assistant subscription, about $20 a month for ChatGPT Plus or Claude Pro, and nothing more is needed.

Put numbers on it. Four articles at five hours each is 20 hours a month. If the manager values their time at $40 an hour, that is $800 of time plus $20 of software. One extra direct booking a month at an average of $900 a week, instead of a booking through a listing site that charges a commission, starts to cover it, and a page that ranks keeps bringing bookings for years without further work.

That can feel slow next to the forty-pages-an-afternoon promise, but four strong articles a month is 48 a year, each built on facts a competitor can't copy. For a small site, that is a meaningful body of work, and every one of those pages also gives AI assistants something specific to quote when a guest asks them for advice.

Questions owners ask about AI-written articles and Google

Should I tell readers an article was written with AI?

Google's guidance suggests explaining how automation was used where a reader would reasonably wonder, and it puts more weight on a visible byline and author background. For a small business article, a short line such as 'Drafted with AI from our booking records, checked and edited by our manager' is honest and costs nothing. What matters more is that a named person has checked every fact.

Do AI detector scores affect my Google rankings?

Third-party detector scores are not part of Google's published guidance, which judges whether a page is helpful, accurate and original. Detectors are also unreliable on edited text and often flag plain human writing. Don't spend time rewriting a good page to lower a detector score; spend it adding first-hand detail, photos and facts a competitor couldn't copy.

How long should each article be to rank?

Google states plainly that it has no preferred word count. Write until the question is fully answered and stop. A focused answer to a narrow question might be 700 words; a buying guide with a checklist and photos might run to 2,000. Padding an AI draft to hit a length target usually makes it worse, because the extra words are the generic ones.

Can I use AI to refresh old articles that have dropped?

Yes, and it is often a better use of time than writing new ones. Paste the old article plus what has changed (new prices, new properties, recent customer questions) and ask the AI to list what is out of date before it rewrites anything. Update the visible date only when the substance has changed, not after a cosmetic edit.

Further reads

• What Is Generative Engine Optimisation and Does Your Business Need It? — What changes when people ask ChatGPT or AI Overviews instead of searching.
• How to Get Your Business Recommended by ChatGPT and AI Search — How to become the business AI assistants actually name.
• How to Check AI Content for Plagiarism Before You Publish — A quick originality check before an AI draft goes live.
• How to Track Whether AI Search Is Sending You Customers — Measure whether AI search is sending you bookings.
• How Much Does Human Editing of AI Content Cost? — What it costs to pay someone for the edit pass.
• How AI Overviews Are Changing Local Search for Small Businesses — How AI Overviews change the clicks a small site gets.
• AI Content Workflow for Agencies: From Brief to Approved Draft — Six gates from client brief to approved draft, with the AI's job and the human's job at each, plus a filled-in brief, outline prompt and claims log.
• Is It Cheaper to Use AI or a Freelancer for Marketing Content? — A per-piece cost comparison of AI and freelance writers for marketing content, counting your own editing time, with three worked cases.
• How to Write Your Website Copy With AI: Home, About, and Services — Page-by-page prompts for your Home, About and Services copy, built from a fact sheet, with sample outputs, a claims check and a 20-minute customer test.
• Does Google Penalise AI-Written Content on a Business Website? — What Google's rules actually say about AI-written pages, how to tell a penalty from a page that never ranked, and a value check for every draft.
• How to Choose an SEO Agency That Understands AI Search — Gather your own AI search baseline, ask agencies the questions that expose GEO hype, score two proposals and know what honest monthly reporting looks like.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Search Central pages on using generative AI content, spam policies (scaled content abuse) and creating helpful, people-first content; Google Search Console Help on the Page indexing and Performance reports.