supereo_
Get started

guide // results // 2026

Do AEO tools work? Every published result, with its sample size

Some changes do, measured against a control. The GEO paper raised a source's visibility by up to 40% in a lab engine against the unmodified source[1][2]. In one published test, re-dated articles gained 31.5% in citations while two unchanged ones fell 10.6%[3]. Other studies found no major uplift once controls were counted, such as a schema study by Ahrefs[4], and several published case studies state no control, or no sample size. supereo is our pick, and we make it: it proves each move against control questions, with sample sizes and ranges on every number. Of the thirteen published results here, ten state a sample size and seven compare against a control.

Updated . Prices checked 7 November 2026.

The verdict

  1. Every move proven against controls

    supereoour pick, and we make it

    • best for teams that want AEO done, not a dashboard: supereo does everything it takes to get you recommended, from your site and pages to the roundups, placements and videos AI reads;
    • it proves each move against control questions, with sample sizes and ranges on every number;
    • run it from Claude or ChatGPT: an MCP server with the same tools supereo's own agent uses.

    From $149 a month

    Run the free auditFree, no card, every answer shown.

Every published result

Published byWhat it foundSampleControlSource
Aggarwal and others, the GEO paperNine rewrite methods, such as adding quotations, statistics and cited sources, raised a source's visibility by up to 40% in a lab engine and up to 37% on Perplexity10,000 queries; the Perplexity check used 200 of the test queriesYes[1] [9] [10]
Puerto and others, C-SEO BenchMost published rewrite methods, the GEO paper's among them, didn't help and often pushed the document down the citation order; adding statistics lowered rankings in 19 of 24 settingsMore than 1.9k queries and 16k documentsYes[11] [12] [13]
Ahrefs, adding schemaPages that added JSON-LD schema got no major uplift in citations on AI Overviews, AI Mode or ChatGPT; a raw 43% rise in AI Mode shrank to 2.4% once control pages were counted1,885 pages that added schema, against 4,000 control pagesYes[4] [7]
Ahrefs, self-promotional pagesA new brand broke into 72 prompt and engine slots it had been absent from, and 82% of those new mentions came in answers citing one of the planted pages34 pages on 5 domains; 9,886 answers from ChatGPT, Gemini, Perplexity and CopilotNo[14] [15] [16]
Otterly, the year in the titleAdding the year to the title and H1 of 11 pages: citations to them rose 56% and 61%, but a page never touched rose 63% over the same time, so the edit gets no credit11 edited pages in two groups, against untouched reference pagesNo[17] [18] [19]
Otterly, a new date onlyChanging only the visible date on 10 articles raised their citations 31.5% in 30 days, while two unchanged control articles fell 10.6%10 re-dated articles and two control articles; 320 prompts on seven AI engines, an estimated 67,200 answersYes[20] [3] [21]
SearchPilot, for OmioAdding brand USPs to pages raised LLM-referred traffic 18% against control pagesnot statedYes[22] [23]
Peec AI, for SIXTNew content on 11 partner listings raised citation frequency 320% on the two fully treated domains, while the unchanged control domain rose 12.2%; the experimenter calls the findings preliminary24 prompts, 11 partner listings and four domains, one of them an unchanged controlYes[24] [25] [26] [27]
Peec AI, for OnBoardA Key Takeaways section raised one page's citation rate 61% over 6 weeks, while an unchanged control page rose about 7%; the page's retrieval ended below where it startedOne changed page and one unchanged page, over 6 weeksYes[28] [29] [30]
Scrunch, for AkamaiAkamai's pages served to AI agents were cited 22% more than its other pages, and brand presence on related non-branded prompts rose 364%not statedNot stated[31] [32]
Profound, for AirbyteVisibility in ChatGPT went from 9% to 26% in one week after Airbyte's content changesOver 500 prompts tracked; the page doesn't tie the result to themNot stated[33] [34]
AthenaHQ, for Cozy EarthCitation volume in its category roughly quadrupled within a quarter of publishing targeted content, while the tracked prompt set kept growingA prompt set that grew from a couple dozen questions to more than 500Not stated[35] [36]
Writesonic, for Blue Star FerriesA 15% visibility lift and 100 new citations in 6 months, after an audit, prompt research and destination guidesnot statedNot stated[37] [38]

Why a result needs a control

AI answers change from one run to the next. One study found that 50–83% of AI citations are gone the next day, and about 75% of those come back within 20 days[5]. In Google's AI Mode, only 9.2% of the exact cited URLs matched across three runs of the same queries[6].

So a before and after can show a gain that was noise. In one schema study, a raw 43% rise in AI Mode shrank to 2.4% once control pages were counted[7].

And a small sample can't tell a real change from chance: with 10 responses per period, a visibility of 75% can't be told from 50%[8].

How supereo proves a move

supereo registers a test before a change ships: target questions the change should move and control questions it can't, asked again and again on one engine before and after. A verdict carries its sample size and range, and a change that made no difference says so.

The method is public, with every number it runs on: How supereo measures.

Have it done: supereo

01supereoour pick, and we make it

  • best for teams that want AEO done, not a dashboard: supereo does everything it takes to get you recommended, from your site and pages to the roundups, placements and videos AI reads;
  • it proves each move against control questions, with sample sizes and ranges on every number;
  • run it from Claude or ChatGPT: an MCP server with the same tools supereo's own agent uses.

From $149 a month

  • Does the worksupereo writes those pages for you, aimed at the questions where AI skips you. supereo.app
  • Proves it against controlsEach change targets a set of questions, and we compare them with a matched set the change can't affect. supereo.app
  • Runs from Claude or ChatGPTRun the same loop from Claude, Cursor or Codex. Your agent gets every tool supereo's own agent uses: your buyers' questions, every answer, the pages AI cites, your brand facts. supereo.app
  • Free auditFree · about 2 minutes · no card supereo.app
PlanA monthPromptsAI engines
Copilot$14940ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity, Claude
Autopilot$49940ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity, Claude

Prices from supereo.app, checked 7 November 2026. What it does, from its own pages, checked 5 November 2026.

How the proof works, with every number it runs on: How supereo measures.

Questions

Do AEO tools work?

Some changes do, measured against a control: in the table, the GEO paper, Otterly's date test and Peec AI's tests for SIXT and OnBoard found gains against a control. Others found no major uplift once controls were counted, such as Ahrefs' schema study. supereo proves each move it makes against control questions.

Which AEO results have a control group?

In the table: the GEO paper, C-SEO Bench, Ahrefs' schema study, Otterly's date test, SearchPilot's test for Omio, and Peec AI's tests for SIXT and OnBoard. The rest compare a before with an after, or with other pages, or don't say.

How big a sample does an AEO test need?

More than most case studies use. With 10 responses per period, 75% visibility can't be told from 50%, and with 50 responses, 50% can't be told from 40%[8]. supereo asks each question again and again, and says how sure each result is.

Why do AEO case studies disagree?

Many are one brand, a few pages and no control, measured over weeks while AI answers move from day to day[5]. A result with a control and a sample size is the one to trust.

How we compiled this

Where this page's words say how an AI engine finds, reads or cites pages, the sentence links the engine's own documentation, or another source, and the day it was read.

Each result is in its publisher's own words under Sources, with the day we read it. The table puts its sample size beside it, or says the source states none, and whether it compared against a control, as the source describes it. A result that doesn't say what it was measured on can't be told apart from noise.

We make supereo, so it is the one tool this page names as a pick.

How supereo measures its own results, with every number it runs on: How supereo measures.

supereo (supereo.app) is an AI agent that does everything it takes to get your brand recommended by ChatGPT, Gemini, Perplexity, Claude and Google, on your site and off it: site fixes, new pages, roundups, placements and video. It proves what worked against control questions. Made by superagnt.

Sources

  1. Through systematic evaluation, we demonstrate that our proposed Generative Engine Optimization methods can boost visibility by up to 40% on diverse queries, providing beneficial strategies for content creators. Among other things, we find that including citations, quotations from relevant sources, and statistics can significantly boost source visibility, with an increase of over 40% across various queries. We also demonstrate the efficacy of Generative Engine Optimization on Perplexity.ai, a real-world generative engine and demonstrate visibility improvements up to 37%. arxiv.org, read 4 November 2026.
  2. We compare them with a baseline, which measures the impression metric of unmodified website sources. We evaluate methods on the complete GEO-bench test split. Further, to reduce variance in results, we run our experiments on five different random seeds and report the average. arxiv.org, read 4 November 2026.
  3. Over the following 30 days, citations to those pages increased 31.5%. Two unchanged control articles moved in the opposite direction, declining 10.6%, from 358 to 320 citations. otterly.ai, read 4 November 2026.
  4. We tracked 1,885 web pages that added JSON-LD schema between August 2025 and March 2026, matched them against 4,000 control pages, and measured citation changes across Google AI Overviews, AI Mode, and ChatGPT. Adding schema produced no major uplift in citations on any platform. ahrefs.com, read 4 November 2026.
  5. Don't judge AI visibility from one check. 50–83% of citations are gone the next day, but about 75% of those come back within 20 days. Count how many of the last seven days you were cited. ahrefs.com, read 4 November 2026.
  6. The average overlap of exact URLs between the three datasets was just 9.2%. This means that AI Mode often chooses completely different pages each time, even when the query doesn’t change. seranking.com, read 4 November 2026.
  7. If we just looked at the raw before-and-after growth of AI Mode, it came in at +43%, but this analysis revealed control URLs gained almost as much, meaning AI Mode was exploding for everyone. Strip that out and the +43% shrinks to the +2.4% shown here. ahrefs.com, read 4 November 2026.
  8. For example, with 10 responses, we can statistically distinguish between 75% and 25%, but not 75% and 50%. With 50 responses, we can statistically distinguish between 50% and 30%, but not 50% and 40%. graphite.io, read 4 November 2026.
  9. To facilitate faithful and extensive evaluation of GEO methods, we propose GEO-bench, a benchmark consisting of 10000 queries from diverse domains and sources, adapted for generative engines. arxiv.org, read 4 November 2026.
  10. We evaluate all our methods on a subset of 200 samples of our test set. arxiv.org, read 4 November 2026.
  11. Our experiments reveal that most current C-SEO methods are not only largely ineffective but also frequently have a negative impact on document ranking, which is opposite to what is expected. Instead, traditional SEO strategies, those aiming to improve the ranking of the source in the LLM context, are significantly more effective. arxiv.org, read 4 November 2026.
  12. This is not a marginal phenomenon: for example, the Statistics method decreases rankings in 19 out of 24 evaluated settings. In product recommendation tasks on Haiku 3.5, 26 out of 30 cases show significant negative effects, and in question answering on o4-mini, 19 out of 30 cases do so. arxiv.org, read 4 November 2026.
  13. In total, our C-SEO benchmark contains more than 1.9k queries and 16k documents. arxiv.org, read 4 November 2026.
  14. After the first pages went live on Feb 13, Evolve broke into 72 of those previously empty slots. Of the 72 new Evolve mentions in prompt-engine slots that had been empty before publishing, 82% appeared in answers that cited one of our pages. ahrefs.com, read 4 November 2026.
  15. We published 34 self-promotional pages featuring our AI visibility tool (Ahrefs Brand Radar) and our conference brand (Ahrefs Evolve) on 5 domains: ahrefs.com, bloggerjet.com, blog.timsoulo.com, medium.com, detailed.com. Self-promotional pages included listicles, reviews, and opinion pieces. We analyzed 9,886 answers from ChatGPT, Gemini, Perplexity, and Copilot. ahrefs.com, read 4 November 2026.
  16. Instead of looking only at total brand mentions, I looked at places where a specific AI assistant, answering a specific query, had not mentioned the brand at all during the Feb 7–12 baseline. ahrefs.com, read 4 November 2026.
  17. We inserted or updated “2026” in the title tag and H1 of 11 pages on our blog, changing nothing else. otterly.ai, read 4 November 2026.
  18. Citations to those pages rose 56% and 61% across two test groups. A page we never touched rose 63% over the same period, so we cannot credit the edit for the growth. otterly.ai, read 4 November 2026.
  19. We call these reference pages rather than a control group, since they were not matched to the test pages on topic, format or citation volume. otterly.ai, read 4 November 2026.
  20. On August 4, 2026, we updated the last published date on 10 existing OtterlyAI articles. We did not change the content itself. otterly.ai, read 4 November 2026.
  21. The experiment covered a US brand report monitoring 320 prompts across seven AI engines, representing an estimated 67,200 AI answers over the initial 30-day period. otterly.ai, read 4 November 2026.
  22. This increased LLM traffic by +18% for Omio. searchpilot.com, read 4 November 2026.
  23. We added clear, well-structured brand USPs to make it easier for LLMs to understand and summarise the page’s value. Our hypothesis was that this would increase the likelihood of these pages being cited or linked to in LLM-driven responses, uplifting LLM sessions on variant pages compared to control pages. searchpilot.com, read 4 November 2026.
  24. Citation frequency on the two full-treatment domains increased 320% (~308% once adjusted for normal background movement), while the partial-treatment domain increased 200% (~188% adjusted). The control domain rose 12.2%, in line with normal AI search fluctuation. peec.ai, read 4 November 2026.
  25. Dr. ElMalah used those sources to shape new content, then tested the approach across 24 prompts and 11 partner listings. peec.ai, read 4 November 2026.
  26. Four domains were included in the experiment: two received the full treatment, one received a partial treatment, and one was left unchanged as a control. SIXT compared a two-week baseline with the two weeks after the changes went live. peec.ai, read 4 November 2026.
  27. Dr. ElMalah describes the findings as preliminary, based on one brand and four domains. peec.ai, read 4 November 2026.
  28. Citation rate climbed at every checkpoint on the test page, from 1.45 to 2.33, a 61% increase over 6 weeks. Absolute citations also increased, from 289 to 387, or 34%. The control page’s citation rate rose by roughly 7% over the same period. peec.ai, read 4 November 2026.
  29. After they added the Key Takeaways section to one page, they left a comparable page unchanged as a control. The test ran for 6 weeks with checkpoints roughly every 2 weeks, and Tyler decided up front that the test would be judged on citation rate, tracked in Peec. peec.ai, read 4 November 2026.
  30. Retrieval told a different story. On the test page, it rose slightly at first, then finished below its starting point. peec.ai, read 4 November 2026.
  31. In terms of citations, AXP pages are outperforming non-AXP pages by 22% since launch. Meanwhile, results for brand presence were significant: AXP pages resulted in a 364% increase in brand presence for associated non-branded prompts. scrunch.com, read 4 November 2026.
  32. The results (comparing AXP pages to non-AXP pages between December 1, 2025 and January 19, 2026) were clear: scrunch.com, read 4 November 2026.
  33. Airbyte's focused strategy delivered rapid, measurable gains: Visibility in ChatGPT tripled in one week, from 9% to 26% tryprofound.com, read 4 November 2026.
  34. Using Answer Engine Insights, Airbyte analyzed over 500 prompts across priority topics in ChatGPT, Perplexity, Google AI Overviews, and Copilot. tryprofound.com, read 4 November 2026.
  35. Within one quarter of publishing its targeted loungewear content, Cozy Earth roughly quadrupled citation volume in the category. Its lead in the loungewear matchup the team was watching grew to 30 points, up from less than one point earlier in the year. athenahq.ai, read 4 November 2026.
  36. Cozy Earth started narrow and expanded deliberately as the program matured, growing the tracked prompt set from a couple dozen questions to more than 500 across bedding, sleepwear, loungewear, and comparison-style queries, against 22 named competitors and five assistants. athenahq.ai, read 4 November 2026.
  37. Blue Star Ferries: 15% visibility lift. 100 new citations. 6 months. writesonic.com, read 4 November 2026.
  38. The agency ran the technical audit, did the prompt research, and built out destination guides on Mykonos, Santorini, and other islands tied to Blue Star's routes. writesonic.com, read 4 November 2026.