Copy for Originals · SEO and AEO
Copy for Originals designs and deploys agentic frameworks that produce original, fact-based content from open source intelligence (OSINT), so Swedish companies can show up both in Google’s search results (SEO) and in the answers from AI services such as ChatGPT, Claude and Perplexity (AEO). The work is done by Jesse Bastide in Tidaholm, Sweden.
Text that says what everyone else says can be produced by anyone in a minute, which is why it is no longer enough to be found. What can rank and get cited is facts of your own: numbers, comparisons and answers nobody else has published. I build a framework in which AI agents collect material from open sources, check every item against its source, and write pages that both Google and AI services can read. A person decides what is true enough to publish. Pricing on request. Send your website and the question you most want to be found for, and I’ll reply in writing with the first thing I’d do.
| Service | Design and deployment of agentic frameworks that produce original, fact-based content for search engine optimization (SEO) and answer engine optimization (AEO) |
|---|---|
| Source material | Open sources (OSINT): official statistics, registers, government pages, forums, reviews and anything else anyone can check |
| Who does the work | Jesse Bastide, with AI agents for collection, sorting and checking |
| Language | Content in Swedish or English. Meetings and calls are in English. |
| Price | On request |
| Meetings | In person in Västra Götaland and Stockholm, remote in the rest of Sweden |
| First step | A free written reply: the first thing I’d do |
| Contact | Contact form or phone +46 70 306 64 82 |
SEO, search engine optimization, is about a page ranking high in Google’s regular search results. AEO, answer engine optimization, is about the page being retrieved and cited when an AI service such as ChatGPT, Claude, Perplexity or Google’s AI Overviews answers a question. One page can do both, but the requirements are not identical.
| SEO | AEO | |
|---|---|---|
| The goal | A high position in the search results | Being retrieved and cited in an AI answer |
| Who reads the page | Google’s crawler, then a person | An AI service’s crawler, which pulls out individual passages |
| What matters most | How well the page answers the search, and how much others trust the website | Every fact in raw HTML, every section understandable by itself, and concrete facts in the text to quote |
| Common mistake | Text that repeats what everyone else has already written | Content that only appears when JavaScript runs, or a firewall that stops the AI services’ crawlers |
The two are less tied together than you might think. According to an Ahrefs measurement reported in our guide to content that LLMs retrieve and cite, only 38% of the pages cited in Google AI Overviews in March 2026 ranked in the top ten for the original query. So a page can be cited without ranking first, if it holds the best answer to a sub-question.
Mainly three things: to be mentioned or cited when someone asks an AI service, to be able to measure whether that happens, and to know what to actually do. That is what our review of 1,073 Reddit threads about AEO from July to October 2026 shows. Of the 811 threads that were not promotion, 42% were about getting cited or recommended, 40% about measurement and 29% about what to actually do.
| What people worry about | Threads | How the question is put |
|---|---|---|
| Getting cited or recommended by ChatGPT and Google’s AI answers | 343 (42%) | “How do you get your site recommended by ChatGPT and Claude?” |
| Measuring AI visibility, and trusting the numbers | 328 (40%) | “How do you actually track AI visibility across ChatGPT and Claude?” |
| What to actually do | 232 (29%) | “So what do we actually need to change in content for AI visibility?” |
| AI answers change from run to run | 98 (12%) | “how do you separate a meaningful trend from ordinary answer volatility?” |
| Is AEO just SEO with a new name? | 87 (11%) | “Is GEO a real discipline or a rebrand to sell new retainers?” |
| Is SEO dying, and where should the budget go? | 82 (10%) | “If you were starting a new business today, would you invest in backlinks or AI search visibility first?” |
| Does it turn into customers? | 75 (9%) | “Has anyone found a reliable way to connect AI visibility with actual business results?” |
| Hiring and paying for AEO services | 61 (8%) | “Is $8k a month fair for "AI SEO" when nobody can show a single citation moved?” |
| The boss or the client is asking about AI search | 49 (6%) | “what exactly are you putting on that slide?” |
| Clicks lost to Google’s AI Overviews | 44 (5%) | “Drop in clicks while impressions grow. AI overview impact or something else?” |
| AI says wrong things about the company | 29 (4%) | “What are you guys doing to influence AI overviews and what AI says about your brand?” |
A thread can belong to up to three rows, so the shares don’t add up to 100. Two things stand out. The most common question isn’t technical. It is about being chosen. And almost as many people wonder whether they can trust the numbers they are given.
Method and limits: 109,824 posts from 14 subreddits were pulled from the Arctic Shift archive. The 1,292 threads from July 1 to October 13, 2026 with at least three comments were read and coded by AI agents against a codebook, and 1,073 were relevant. 248 of those (23%) promoted a tool or a service and are not in the table. 40 threads were checked by hand, and when the same 60 threads were coded three times the main category matched in 44 to 49 cases, so the counts should be read as approximate. Most posters don’t say who they are. The review was produced with the framework described further down.
Partly. Google writes in its guidance that optimizing for generative AI search is still SEO, and for Google’s own AI features that is broadly true. But Google speaks only for Google Search. ChatGPT, Claude and Perplexity fetch and read pages in their own way, and there it isn’t enough for the page to rank. The sensible course is to follow Google where independent measurements back Google up, and to go further where the mechanics of the AI services call for it.
“From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.”
Google Search Central, guide to generative AI features on Google Search, last updated July 10, 2026
Note the first words: from Google Search’s perspective. The guide describes how Google Search works and what Google wants site owners to do for Google. It says nothing about how other AI services fetch pages. Here is how Google’s advice holds up against what can be measured:
| Question | What Google says | What independent evidence shows | Our conclusion |
|---|---|---|---|
| llms.txt | Google Search ignores the file | Ahrefs found that 97% of the llms.txt files on 137,210 domains received no requests at all in May 2026 | Google is right. We spend no time on the file. |
| Schema markup | Not required for AI search | Ahrefs found no reliable increase in citations for 1,885 pages that added markup | Google is right about citations. Markup should mirror visible text and never be the only place a fact appears. |
| Splitting text for AI | No requirement to break content into tiny pieces | The crawlers from OpenAI, Anthropic and Perplexity don’t run JavaScript, and retrieval systems judge one passage at a time. Anthropic cut failed retrievals by 49% by adding the context back to each passage. | Here we go further than Google. The text isn’t chopped up, but every section has to make sense alone and every fact has to be in raw HTML. |
| What matters most | Content that is unique and goes beyond common knowledge | In the GEO study, statistics, quotations and cited sources raised visibility by up to about 40%. In 252,000 controlled trials across six language models, explicit prices and recent dates increased selection. | Agreed. It is the whole service. |
| Tools that measure “AI ranking” | Third-party tools have no access to Google’s data and can’t guarantee performance | In SparkToro’s study, the same prompt returned the same list of brands in fewer than 1 in 100 runs | Agreed. We promise no position in AI answers. |
The reason to go further than Google’s advice is mechanical. An AI service fetches the HTML, extracts the main text, splits it into passages and quotes the passages that best answer the question. A page written for that chain also works for Google. The reverse isn’t always true. The whole chain, with sources, is described in our guide.
On Reddit, the most established voices almost all agree with Google. Among moderators and flair-verified professionals in the SEO forums, 23 of 24 coded replies said AEO is the same work as SEO. Across all 2,969 coded top replies it was closer: 151 to 104. The question they rarely answer was asked by one poster: “If AEO and GEO is 90% SEO then why is my website ranking on google but not showing in AI search?” The answer is that ranking and being cited are two different steps.
One reservation: the evidence is strongest for the mechanics and for controlled experiments. Case studies of individual companies’ results mostly come from the people selling the tools, and we don’t cite those.
Nobody can order a mention, but three things have to be in place for one to be possible: the AI services’ crawlers must be allowed to fetch the page, the page must contain a clear answer with facts that can be quoted, and the website must be in Google’s and Bing’s indexes. Beyond that, what others write about the company matters, and no company fully controls that.
The service covers the first two points. Mentions planted for the AI services’ sake are not part of it.
A company doesn’t show up on Google or in AI answers for one of three reasons: the website has no page that answers the question customers ask, the page says the same thing as a hundred others, or the crawlers can’t read it. The first and second are solved with content, the third with technology.
The fear of being invisible online is reasonable. Someone who asks an AI service about a supplier in your industry gets an answer built from the pages the service could read. If your page isn’t among them, you never learn that the question was asked.
Measure what the search services report themselves and what shows up in your own systems, and be careful with tools that state a position in AI answers. Since 2026 Google Search Console has a report on impressions in AI Overviews and AI Mode, and Bing Webmaster Tools shows how often pages are cited in Copilot. In SparkToro’s study, the same prompt returned the same list of brands in fewer than 1 in 100 runs, so a single position says almost nothing.
| Source | What it shows | Limit |
|---|---|---|
| Google Search Console, the generative AI report | Impressions in AI Overviews and AI Mode, by page, country and date | Clicks are not reported. The report has been available to all websites since August 31, 2026. |
| Bing Webmaster Tools, AI Performance | How often your pages are cited in Copilot and in Bing’s AI summaries, and which pages | Bing says the query data is a sample |
| Your own web analytics | Visits that arrive through a link in an AI answer | Only those who click show up, not those who read the answer and moved on |
| Ask the customer | “How did you find us?” in the form or on the call | Someone has to ask, and write the answer down |
| Third-party tracking tools | How often a brand is mentioned across a set of prompts | Answers change between runs. Look at how often you are mentioned across many runs, never at a position. |
On Reddit, measurement is the topic where promotion is densest. Of 213 top-scored replies recommending a tracking tool, only 76 were not themselves promotional, and among the most established voices 15 replies doubted the tools’ numbers while 3 defended them. For now I don’t sell reports on position in AI answers.
That is not settled. What can be measured is that Google’s own AI answers reduce clicks: Ahrefs measured a 58% lower clickthrough rate for the top-ranking page when an AI Overview was shown, in data from December 2025. Whether visitors from AI services become customers is disputed. In our coded Reddit replies, 29 said they convert well and 31 said the volume is too small to matter.
The conclusion is not to move the budget from Google to the AI services. It is that the same page should work for both, and that you should ask new customers how they found you, so you know instead of guessing.
Original content built on open sources is pages that report something new, worked out from publicly available material such as statistics, registers, government pages, forums and reviews. OSINT stands for open source intelligence. The point is that the reader can check every item, and that no competitor has the same numbers.
The difference from ordinary content production is where the text starts. An ordinary article starts with a keyword and summarizes what is already online. A page built on open sources starts with a question customers ask that nobody has answered with numbers: What do companies in our industry actually pay? How many of our competitors do this? What do customers complain about most?
That kind of material has the strongest support in the research on what AI services cite. In the GEO study (KDD 2024), reported in our guide, statistics, quotations and cited sources raised a source’s visibility in generated answers by up to about 40%, while keyword stuffing did worse than doing nothing.
The agentic framework splits the work into fixed phases and gives each phase to whoever does it best: AI agents collect, sort and check the material, and a person picks the question, judges the conclusions and approves what gets published. The phases are the same every time, so the result can be audited afterwards.
| Task | Who |
|---|---|
| Choosing the question and the sources | A person |
| Searching, fetching and sorting the material | AI agents |
| Comparing quotes with the source | A script and independent grader agents |
| Deciding which conclusions hold | A person |
| Approving what is published | A person: me and you |
Not without checking, which is why checking is a phase of its own. In our own study of nine agent runs, a script found all 593 quotes word for word on the pages given as sources, but two independent graders judged that 41 of the 593 claims (6.9%) said more than the quote showed. The quotes were real. It was the conclusions that went too far.
That is how AI-produced content usually goes wrong: the source exists, but it doesn’t quite say what the text claims. A framework that only checks that the link works misses it. So every claim is compared with its quote, and a person decides what stays. The full study is here: Haiku, Sonnet or Opus for online research?
No. Traditional programmatic SEO fills a template with a list of keywords and produces hundreds of pages that say the same thing with different place names. An agentic framework for original content does the opposite: one question, new material, one page. AI is used as an intelligent collaborator that carries out tasks hands-off and coordinates other agents, not as a machine that prints pages. Google’s scaled content abuse policy is about many pages with no value to users, no matter how they were created.
| Traditional programmatic SEO | Agentic framework for original content | |
|---|---|---|
| Starting point | A keyword list times a template | A question customers ask |
| Source material | What is already online, reworded | New material, collected from open sources |
| Number of pages | Hundreds or thousands | One page per question |
| Checking | None, or spot checks | Every quote compared with its source, and a person approves |
| What the reader gets | The same answer as on every other page | Facts that exist nowhere else |
Google’s own texts draw the line at value, not at the tool. The scaled content abuse policy describes large amounts of unoriginal content “that provides little to no value to users, no matter how it’s created”. Google’s guidance on generative AI also says the technology “can be particularly useful when researching a topic, and to add structure to original content”, and that all AI-generated content should be fact-checked by a person before it is published. That last part is the checking phase and the approval in the framework.
The line cuts both ways. An agentic framework can also be misused to print pages with nothing new in them. That is why the checking and the approval are not optional.
Yes. The studies on this website were produced with the same method: one question, open sources, a dataset and a page where every number can be traced. They are about American markets, but the method is the same for a Swedish company.
| Study | Source material | One result |
|---|---|---|
| Lead generation at 100 Chicago law firms | Google local results, Google’s ad register, the firms’ websites and reviews | 48 of 100 firms have ads on record with Google |
| Do websites block AI crawlers? | robots.txt, llms.txt and response codes from 100 websites | 13 of 78 sites refused a request carrying an AI crawler’s name |
| What do marketing agencies charge? | 312 coded forum threads from 2023 to 2026 | Owners’ own reports of fees, one figure per owner |
| Why did Google Ads cost per lead rise in 2026? | 3,173 forum posts, of which 123 coded first-hand threads | 31 of 123 posters (25%) blamed Google’s August 17 change |
| Which AI model does the best research? | Nine agent runs, 593 checked quotes | All 593 quotes were found word for word on the source pages |
Pricing is on request. The cost depends on how many questions are to be answered, which sources have to be collected and whether the content should exist in Swedish, English or both. The first written reply, on what I’d do first, is free.
If I also run your Google Ads campaigns, the work connects: the search terms in the ad account show which questions customers actually ask, and those are the first questions worth answering with a page of their own. See Google Ads lead generation in Sweden.
This is not a fit for companies that want a guaranteed ranking, many texts fast or a new website. Nobody can promise a position on Google or a mention in an AI answer, and anyone who does is guessing.
No. The AI agents collect, sort and check the material and write drafts. Which question to answer, which conclusions hold and what gets published is decided by a person.
No. Nobody can promise a position on Google or a mention in an AI answer. What I promise is that the content is built on facts that can be checked and that the page is written so search engines and AI services can read it.
Swedish or English. English is my native language. Swedish copy is produced with the AI tool Claude and proofread by Swedish-speaking proofreaders at Copy for Originals before it is published. The Swedish version of this page was produced the same way.
Ranking and being cited are two different steps. According to Ahrefs, only 38% of the pages cited in Google AI Overviews in March 2026 ranked in the top ten for the original query. ChatGPT, Claude and Perplexity also fetch pages with their own crawlers, which can be shut out by a firewall and which don’t run JavaScript.
Ask what is delivered, how it is measured and what happens if nothing changes. Warning signs are guaranteed mentions, a promised position in AI answers, llms.txt as the main deliverable and a large number of pages a month. Google writes that third-party tools have no access to Google’s data and can’t guarantee performance. We have not found any independent survey of what AEO services cost.
That three things can be shown without guessing: whether the AI services’ crawlers can reach the website, how many impressions the pages get in Google’s AI answers according to Search Console, and which customer questions the website has no answer to. That is a more honest picture than a position in an AI answer that changes the next time the question is asked.
Not directly, because no outsider can edit an AI answer. What can be done is to publish the correct facts clearly on your own website, in raw HTML and with dates, so the service has something correct to fetch the next time the question is asked. There is no guarantee the answer changes.
Only if your customers actually ask questions there and you have something factual to answer. In our coded Reddit replies, third-party mentions were named most often as the way to get cited, but Google writes that seeking inauthentic mentions isn’t as helpful as it might seem. About one in four threads about AEO in our review was promotion, and it tends to be recognized.
Not measurably, according to the studies reported in our guide. SE Ranking found no relationship between llms.txt and how often a domain is cited across about 300,000 domains, and Ahrefs found no reliable increase for 1,885 pages that added schema markup. What has support is facts in raw HTML, sections that can be understood alone and concrete facts to quote.
Yes. Copy for Originals also manages Google Ads for lead generation. The ads give visibility as soon as the campaign is running, and the content stays when the ad budget is switched off.
Meetings and calls are in English. Jesse Bastide is American and a native English speaker.
Paste the address of your website and write the question you most want customers to find you through. I’ll reply in writing with the first thing I’d do. No charge, and you don’t owe me a call.
The form is on copyfororiginals.com. Phone: +46 70 306 64 82. Meetings and calls are in English.
Related: Google Ads lead generation in Sweden and English copywriting for Swedish companies selling to the US.
Sources were read on October 13, 2026. Figures attributed to our guide have their own sources there. The Reddit figures come from our own review, described under the table above.