SEO

AI for SEO: what works and what Google penalises

The tasks where using AI for SEO saves hours, two questions to decide what to delegate, the mistakes that cost you rankings and a workflow with human checkpoints your team can apply to the very next page.

AI for SEO: where it helps and where it doesn't

The marketing manager of an online wine shop in Bristol has three hundred product pages with no meta description and a free afternoon. With ChatGPT she could have them all drafted before she goes home. What she can't tell is whether AI for SEO, used this way, helps her rankings or puts them at risk.

Bringing AI into SEO work is now routine in many small businesses, and the answer depends on the task. AI performs well on analysis, sorting and first drafts, where a professional can check the output in a few minutes. It performs badly when asked to supply facts, market data or experience it does not have. A language model with no access to a data tool has no idea of the real search volume for a keyword, or of the clicks your site receives.

Google's position points the same way. Its systems reward content that helps people, however it was produced, and its spam policies target pages built to manipulate rankings. The risk starts when whatever comes out of the tool goes live without anyone reviewing it.

This matters more now that some searches are answered with AI-generated summaries, such as Google's AI Overviews. The pages those summaries cite tend to be the ones that answer clearly with reliable facts, which is exactly what a workflow with human review protects.

Which SEO tasks to hand to AI: three kinds of work

Before you pass a task to AI, ask two questions. They sort any SEO job into one of these three groups.

  • Quicker to check than to do

    Grouping five hundred searches, suggesting titles or generating structured data markup. Reviewing the output takes minutes, and doing it by hand would take hours. Delegate with confidence.

  • Needs data the model doesn't have

    Volumes, clicks, rankings or prices. AI can analyse them, but only if you give it the export from Search Console or your SEO tool. If it answers without that data, it is making it up.

  • Depends on business judgement

    Which services to prioritise, which pages bring in customers, what a page promises. AI can lay out the options, but the decision belongs to someone who knows the margins, the customers' questions and the team's capacity.

Six SEO tasks where AI earns its place

In each of these, AI does the heavy lifting and a person on your team can check the result quickly.

  • Keyword and intent research

    Ask AI to group a list of searches by intent, meaning what the searcher wants (to learn, compare or buy), and to suggest the phrasing your customers actually use. Volumes and difficulty come from Search Console or an SEO tool.

  • Content briefs and outlines

    From the keyword and your customers' real questions, AI drafts a brief with the page's goal, the reader, the sections and the FAQs. The writer spends their time on what only they know.

  • Internal linking ideas

    Give AI a list of your URLs and page titles and it will suggest which pages should link to each other, and with what anchor text. An accounting firm might find that its guide to VAT returns never links to its tax services page.

  • Draft titles and meta descriptions

    For a large catalogue, AI proposes a title and description for each page at the length you set. Someone then checks that each one describes the product accurately and that the same phrases are not repeated everywhere.

  • Structured data

    AI writes JSON-LD markup, code that describes the page to search engines, for a product, an article or an FAQ in seconds. Validate it with Google's Rich Results Test and check that every value matches what visitors can see on the page.

  • Technical audits and Search Console

    Export the Performance report or a site crawl and ask AI to spot patterns, such as pages with many impressions and few clicks, duplicate titles, broken links or queries that have dropped. You decide what to fix first.

Common AI for SEO mistakes that cost rankings

Nearly every problem comes from skipping review or asking AI for something it cannot know.

  • Unedited mass content

    Publishing dozens of articles in a week that look alike and repeat what the top results already say. Google calls this scaled content abuse, and the pages involved can drop in rankings or vanish from results.

  • Invented facts

    Language models state figures, studies, rules and dates that do not exist with complete confidence. A false figure damages the reader's trust. Check every fact against its original source.

  • Structured data that doesn't match the page

    A price, rating or question marked up in the code that visitors cannot see on the page goes against Google's guidelines.

  • Copy-paste meta descriptions

    Bulk-generated text that repeats the same formula across hundreds of pages. Google may ignore it and show a different snippet, and searchers get no reason to pick your result.

  • Handing over priorities

    Let the tool choose the topics and your site will end up covering the broadest searches in your sector, which are often not the ones your customers make.

An AI for SEO workflow with human checkpoints

Use it to create a new page or improve an existing one. Each step says what AI does and what a person signs off before moving on.

  1. 01

    Set the page's goal

    A person decides which page is needed, for which customer and what it should lead them to do, for example getting a homeowner to request a valuation. Do not delegate this step.

  2. 02

    Research with AI, validate with real data

    AI groups searches by intent and suggests questions. Someone checks the list against Search Console and a keyword tool, and drops anything with no demand or no fit with your business.

  3. 03

    Write the brief

    AI drafts the structure, the FAQs and suggested internal links. Your content lead corrects it and adds what competitors cannot say, such as a real case or an internal figure.

  4. 04

    Draft and enrich

    The draft can come from AI or from a person. Either way, an expert in your company reviews it, checks every fact and adds their own experience. Without that review, the page does not go live.

  5. 05

    Prepare the technical layer

    AI proposes the title, the meta description and the structured data markup. Someone checks the length and the consistency with the visible content, then validates the code.

  6. 06

    Publish and measure

    Follow the page in Search Console for several weeks. AI can summarise how clicks, impressions and queries change. A person decides whether to expand, correct or merge the page.

Example

The Bristol wine shop starts with its low-CTR pages

Back to the Bristol wine shop, an illustrative case. Instead of rewriting all three hundred pages at once, the manager exports the last three months of page data from Search Console and asks AI to flag the pages with many impressions and a low CTR. CTR, or click-through rate, is the share of times your result appears and someone clicks it.

She picks about forty pages from that list. For each one, AI suggests a title and meta description with the producer, the region and a typical use, such as a wine for Sunday roast. The shop's wine buyer reviews every text, corrects a vintage the AI got wrong and strips out repeated phrases. AI also generates the product markup, and the manager checks that each price matches the page.

To measure it, she leaves another forty pages with similar impressions untouched. After a few weeks she compares CTR and clicks for the two groups in Search Console. If the edited pages improve more than the others, she rolls the method out to the rest of the catalogue. If not, she reviews the texts before going further.

Your first work list, straight from Search Console

Open Search Console, export the Performance report by page and paste it into your AI assistant. Ask for the ten pages with the most impressions and fewest clicks, with a short hypothesis for each. Check three of them yourself. If the hypotheses make sense, you have your first work list. If not, you have learned where AI needs more context.

Obinova is an AI agency based in Barcelona and Paris. Our training teaches your team to use ChatGPT, Claude, Gemini and Microsoft Copilot on real SEO tasks, such as grouping keywords, preparing a brief, reviewing a draft with a critical eye and analysing a Search Console export. It runs on a dedicated video platform or in person at your company.

Our AI audit and consulting looks at how your marketing and content team works today. Together we identify the tasks where AI saves time, the ones that should stay with a person and the checks worth adding. A mission starts with a launch kick-off, followed by follow-ups several times a year.

Frequently asked questions

Can AI replace an SEO specialist?

It can speed up part of the job, such as grouping keywords, drafting metadata or reviewing data exports. What it cannot do is know your business, set priorities or check that information is correct. In practice, a specialist who uses AI well covers more ground in the same time, and their judgement is still needed for every decision.

Is ChatGPT good for SEO, or do I need other AI tools?

ChatGPT is useful, and so are Claude, Gemini and Microsoft Copilot. These general assistants handle most writing, sorting and analysis tasks. For demand and performance data you need reliable sources, such as Google Search Console and a keyword research tool. Many SEO tools now include AI features of their own. When choosing between them, look first at the quality of their data.

Does Google penalise meta descriptions written with AI?

Google judges them on quality, whatever their origin. Each description should sum up the page well and give people a reason to click. If Google decides it does not match the search, it may show a different snippet. Watch above all for descriptions repeated across many pages, which is common when they are generated in bulk without review.

Can I use AI to generate structured data?

Yes, and it is one of the tasks where AI saves the most time. Ask for JSON-LD, the format Google recommends, and validate it with the Rich Results Test. Check that every property matches the visible content. Marking up ratings, prices or questions that visitors cannot see on the page goes against Google's guidelines.

How much time does using AI for SEO save?

It depends on the tasks and on how your team is organised, so it is worth measuring in your own case. Note how long it takes today to prepare a brief, review the metadata for a section or analyse Search Console. Compare after a few weeks of working with AI and the workflow above. You will see where you gain time without losing quality.

Does AI for SEO also help you appear in ChatGPT or AI Overviews?

The foundation is the same. Assistants that search the web need crawlable pages with clear answers and reliable facts, which is what a sound SEO workflow with human review produces. What changes is measurement, because AI answers vary between sessions, so it is worth asking the same questions every month and watching the trend.

Put AI to work on your SEO

Book a free 30-minute video call. We will look at the SEO tasks your team handles today, which ones you can safely hand to AI and which training or audit fits best.

Free 30-minute diagnosis →
Free diagnosisWhatsApp

WhatsApp · Reply within 2 hours during business hours