Using AI for SEO works if it produces two pieces of verifiable evidence: more clicks from Google on the pages you've published and more citations of your brand in answers from ChatGPT, Gemini and Perplexity. Everything else is a side dish. Checking them takes just one hour a month.
AI-powered SEO is the use of AI to plan, write and optimize content that ranks on Google and gets cited by AI answer engines. Here you'll find the process for measuring it without inflated scores: four steps, the mistakes to avoid and a complete example.
Key takeaways
- The only two pieces of evidence that matter are the clicks recorded in Google Search Console and the number of AI answers that name your brand.
- A fixed list of 10 questions, asked every month to 3 engines (ChatGPT, Gemini, Perplexity), produces 30 answers you can compare over time.
- Google doesn't penalize a text because it was written with AI: it penalizes mass-produced content that offers no value to the reader.
- Rising impressions with flat clicks almost always point to a title or meta description problem, not a content problem.
- A tool's "SEO score" is not a result: no customer ever came from a 92/100.
What does it mean for AI-powered SEO to work?
It works when pages written with AI get real clicks from Google and when AI answer engines cite your brand or your site for your customers' questions. These are two different channels: classic SEO aims for the blue links, GEO aims to be the source inside a generated answer.
You need both pieces of evidence because they measure different things. A page can get zero clicks and still show up as a source in Perplexity. Conversely, a page can bring in traffic from Google without ever being named by ChatGPT. If you only look at one, you see half the picture.
There's also a question many business owners ask before anything else: does Google punish text written with AI? No. Google made this clear in its official position on AI-generated content, published in February 2023: what counts is usefulness to people, not how the text was written. The real risk is publishing lots of identical, empty texts to manipulate rankings. So the right measure isn't "was it written by AI?" but "does it bring results?".
Step 1: which pages do I measure in Google Search Console?
Measure only the pages published with the help of AI, one by one, in the Performance report in Google Search Console. Compare the last 28 days with the previous 28 and note four numbers per page: clicks, impressions, CTR and average position. Time needed: about 20 minutes.
Follow this sequence:
- Open Search Console and go to Performance > Search results.
- Turn on all four metrics: total clicks, total impressions, average CTR, average position.
- Set the date filter to Compare, last 28 days against the previous 28.
- Open the Pages tab and filter by URL or by folder (for example
/blog/). - Export to a spreadsheet and keep only the pages that came out of your AI workflow.
Two things to watch. First: a newly published page needs weeks before it shows stable data, so flag pages less than two months old and don't judge them yet. Second: if you sell in several countries, add the Country filter. A German-language page that grows in Austria and stays flat in Germany tells a specific story.
Don't stop at the site total. Overall traffic can rise thanks to two old pages and hide the fact that the twenty new ones bring in nothing. The question is always per page. If you've never done a general check of your site, start with a website SEO analysis before measuring individual pieces of content.
Step 2: how do I prepare the 10 questions to ask AI tools?
Write 10 questions a real customer would ask before buying, in natural language, and freeze them: you'll use them word for word every month. Mix 4 general category questions, 3 comparison questions, 2 questions about the customer's problem and 1 question with your brand name. Time needed the first time: 15 minutes.
Take the questions from real sources, not from your imagination. Search Console queries, customer emails and the questions your sales team hears every week are a gold mine. A good question sounds like a person, not like a keyword.
| Type | How many | Example (management software company) |
|---|---|---|
| Category | 4 | "What's the best management software for a small practice?" |
| Comparison | 3 | "Is cloud-based or installed management software better?" |
| Problem | 2 | "How do I cut the time wasted on invoicing?" |
| Brand | 1 | "What do you think of [your brand]?" |
The question with the brand name serves a different check: it tests what AI tools say about you, and whether they say it correctly. The other nine measure whether you get cited when nobody names you. These are the ones that count most.
If you work across several markets, write the 10 questions directly in each market's language. Don't translate them word for word: a Spanish customer doesn't phrase a question the way an Italian one does. A clumsy translation gives you a flawed test.
Step 3: how do I log brand citations each month?
Ask the 10 questions to ChatGPT, Gemini and Perplexity in a new chat, and for each of the 30 answers note three things: whether the brand is cited (yes/no), whether your site appears among the sources (yes/no), and which competitor is named. Your citation rate is the number of "yes" answers divided by 30.
Always use a new conversation, with no history. AI tools take previous chats into account and may "remember" you in a distorted way. Copy the question exactly as written in the spreadsheet, without tweaks.
The simplest log is a spreadsheet with one row per question and one column per engine:
| Question | ChatGPT | Gemini | Perplexity | Competitor cited |
|---|---|---|---|---|
| Q1 category | No | Yes | Yes (source) | Competitor A |
| Q2 category | No | No | No | Competitor A, B |
| ... | ... | ... | ... | ... |
Perplexity shows its sources explicitly, so there you can see right away which of your pages was used. ChatGPT with web search turned on does the same. When an answer cites one of your pages, note the URL: it's the direct link between step 1 and step 3.
AI answers change from one session to the next. Don't be fooled by a single month. Look at the trend over three months: going from 3 to 9 citations out of 30 is a real signal, going from 6 to 7 is noise.
Step 4: what do I fix when a page isn't growing?
The fix depends on the symptom: zero impressions points to an indexing problem, high impressions with few clicks point to a weak title, a position between 8 and 20 points to content that needs strengthening. No AI citations with good Google traffic usually means answers that are hard to extract. Time: the remaining 25 minutes.
| Symptom | Likely cause | Fix |
|---|---|---|
| Zero impressions after 2 months | Page not indexed or blocked | URL Inspection in Search Console, request indexing, check noindex tags and sitemap |
| Rising impressions, flat clicks | Unappealing title and meta description | Rewrite the title around the concrete benefit, test for 4 weeks |
| Average position 8-20 | Content less complete than competitors' | Add examples, verifiable data, a practical case written by someone who knows the subject |
| Google traffic fine, zero AI citations | Answers buried in the text | Open each section with a direct 40-60 word answer under a question heading |
| AI cites only competitors | Little brand authority outside your site | Seek mentions on authoritative third-party sources, clear structured data |
Fix one thing at a time. If you change the title, text and structure on the same day, next month you won't know what worked. Write the date of every change in the spreadsheet.
The most expensive mistake here is regenerating the whole page from scratch with AI. You lose the parts that were already working. A targeted fix with human review is better: AI prepares the draft, a person adds first-hand experience and real cases and checks the facts. If you use WordPress, also check the technical basics with the SEO steps for WordPress before blaming the content.
Which metrics are just for show and should be ignored?
Ignore any number a tool calculates on its own without tying it to real clicks or citations: the "SEO score" out of 100, the "content score", keyword density, the number of articles published and "AI visibility scores" that don't show the underlying answers. They're theatre, not results.
The problem isn't that they're always wrong. It's that you can't verify them. A 94/100 tells you the page meets the tool's rules, not Google's or a customer's. And many tools push you to write to raise the score, with longer, more repetitive texts.
Here are the most common traps:
- Publishing volume. Fifty articles a month with no clicks are worth less than five that bring in enquiries. Mass-producing similar texts is exactly what Google's scaled content abuse policy targets.
- Number of keywords "ranking". Sitting at position 74 for 300 keywords brings nobody to your site.
- Opaque AI visibility scores. If the tool doesn't show you the exact question and the exact answer, you don't know what you're measuring.
- Total site traffic. It hides which pages are working and which aren't.
If your toolbox is full of dashboards showing these numbers, it's worth clearing out your SEO tools and keeping only the ones that lead to a decision.
Example: what does a full monthly check look like?
Let's take a hypothetical case: an office furniture company that has been publishing 8 articles a month with AI for three months, in Italian and German. The October check takes 60 minutes and leads to three concrete fixes. The numbers are illustrative and are there to show the reasoning.
Minutes 0-20, Search Console. Of 24 articles, 6 are less than two months old and get set aside. Of the remaining 18, 5 bring in almost all the clicks. An article on ergonomic chairs has thousands of impressions but a very low CTR. Three German articles have zero impressions.
Minutes 20-40, AI questions. The 10 questions produce 30 answers. The brand appears in 4: three times in Perplexity, once in Gemini, never in ChatGPT. The same competitor is cited in 7 answers, and it has a comparison page with a table and indicative prices.
Minutes 40-60, fixes.
- Rewrite the title of the ergonomic chairs article around a specific benefit.
- Inspect the three German URLs: two turn out to be excluded because of a noindex tag left over from the draft.
- Add a comparison section with a direct answer at the top to the strongest article, modelled on what AI tools already cite.
October citation rate: 4 out of 30. Realistic goal for January: grow steadily for three months, not double in one. Everything logged in the spreadsheet, with the date of each change.
That's the point of the method. Every number leads to an action. If you want to know when these clicks turn into a financial return, you can calculate your SEO break-even point starting from the data in your check.
Frequently asked questions
Do I have to disclose that content was written with AI to avoid penalties?
Google doesn't require a "written with AI" label to rank a page. What counts is that the text is useful, accurate and written for people. Stating who reviewed the content does, however, strengthen the reader's trust. The risk of a penalty only arises if you publish large volumes of texts with no added value to manipulate rankings.
How long should it take for a page written with AI to show results?
It depends on the site's authority: a domain with a solid history sees impressions sooner than a new one. Only judge a page after at least two months of data in Search Console. If impressions are still zero after that period, the first thing to check is indexing, not rewriting the text.
What do I do if an AI says wrong things about my brand?
Note the incorrect answer and the source, if the engine shows it. The error often comes from an old page on your site or a third-party site with outdated information: fix that source. Then publish a clear page on your site with the correct facts (prices, services, markets) and recheck the same question over the following months.
Can I use the same 10 questions for every market?
You can use the same structure (4 category, 3 comparison, 2 problem, 1 brand), but not the same wording. Each market has different competitors, habits and phrasing. Keep a separate spreadsheet per language, so a drop in Germany isn't hidden by growth in Italy.
Schedule your first check by Friday
Open your calendar and block out one hour on the first Monday of every month. Today, prepare just two things: the per-page Search Console export and your 10 questions. Next month you'll have your first real comparison.
If you'd rather start from a ready-made diagnosis, UPGeoSEO offers a free SEO + GEO audit of your site, with a consultant in chat who guides you through the fixes. The Audit only plan costs $9, monthly or one-off. With Autopilot ($40/month) we plan, write natively in every language and publish to your CMS, and we monitor your visibility in AI answers. You approve, we publish. And you measure everything with real clicks and citations, not vanity scores.
