AI Search Optimization: Insights from 41M Results (2026)

AI search optimization improves the chance that a brand or page is mentioned or cited in an AI-generated answer. The 2025 Profound study showed that citation sources differed sharply by platform. In SEOmator's July 2026 snapshot, discovery-prompt mention rates still ranged from only 11% to 18%, so visibility must be measured per engine and query.

Key findings:

• Profound analyzed more than 41 million results across ChatGPT, Google AI Overviews, Perplexity, and Microsoft Copilot.
• A 650-execution test found only 12% overlap between ChatGPT results and Google results, and 26% overlap with Bing.
• Traffic and backlink counts had weak bivariate relationships with citations in a separate 50,000-prompt analysis.
• Comparative listicles accounted for 32.5% of 177 million categorized citations, but that descriptive share does not prove list format caused the citations.
• Across SEOmator's July 2026 category prompt sets, the top three brands captured about 58% of all AI mentions.

The original article summarized Josh Blyskal's BrightonSEO presentation. This update preserves the useful findings, links them to the original slide deck, and checks the strongest recommendations against official 2026 platform guidance and SEOmator's own prompt-tracking data.

At Brighton SEO 2025, Profound's AI Search Strategist Josh Blyskal revealed findings from their analysis of over 41 million AI search results across ChatGPT, Google AI Overviews, Perplexity, and Microsoft Copilot. The data paints a clear picture: AI search is not only legitimizing its position but is fundamentally different from traditional search, creating both challenges and opportunities for SEO professionals.

This technical analysis breaks down the key findings and provides actionable insights for optimizing your visibility in AI search.

What did the 41M-result study actually measure?

The study combined several analyses, not one uniform 41-million-row experiment. That distinction matters because each headline finding has a different sample, query set, and question it can answer.

At BrightonSEO in April 2025, Profound's AI Search Strategist Josh Blyskal presented findings from an analysis of more than 41 million AI search results across ChatGPT, Google AI Overviews, Perplexity, and Microsoft Copilot. The official session listing retained the earlier "10 million" title, while the published slides state that the final analysis exceeded 41 million results.

The query mix included commercial and informational intent. The slides don't publish enough sampling detail to treat every percentage as a universal benchmark for every market, language, or prompt type. They do provide a useful 2025 snapshot and a set of hypotheses worth testing against your own queries. SEOmator's 2026 comparison data comes from a separate panel of 1M+ customer prompts and keywords tracked each month; we use it to test the presentation's direction, not to retroactively change Profound's methodology.

How different were AI citations from traditional search?

The first and most critical finding: AI search cannot be the same as traditional search. The very nature of conversational AI interfaces creates a substantially different user experience compared to the familiar 10 blue links.

When comparing search results between platforms, the data reveals minimal overlap:

• ChatGPT results overlap only 12% with the Google SERP, based on analysis of 650 individual ChatGPT executions
• ChatGPT results overlap only 26% with Bing, despite ChatGPT's browsing functionality being powered by Bing

This minimal overlap confirms that optimizing solely for traditional search engines will not guarantee visibility in AI search results. It does not show that Google or Bing rankings are irrelevant. Retrieval systems, indexes, query rewriting, model choices, and citation selection can each change the final source set.

One of the most significant findings from the research is a fundamental shift in the relationship triangle between users, content, and search engines. In traditional search, users interact directly with websites. In AI search, the AI acts as both connector and arbitrator, owning the direct relationship with users.

In the past, search engines simply connected users to websites, allowing for direct interaction. The current reality is dramatically different:

• The AI now sits between users and content sources
• Users develop a relationship with the AI, not with content providers
• The AI determines which content is relevant and how it's presented
• Content providers must appeal to the AI, not just to users

This changes the unit of success. Traditional SEO measures rankings and clicks. AI search also requires query-level measurement of mentions, cited URLs, citation context, and share of voice. Our AI-powered SEO comparison covers where the two disciplines still share the same foundations.

Which sources did each AI platform prefer?

Each platform showed a distinct source mix in the 2025 sample. These patterns are useful for planning distribution, but they are snapshots rather than permanent ranking rules. Models, retrieval partners, indexes, and answer formats change.

ChatGPT source preferences

ChatGPT heavily favors Wikipedia (1.3M citations), followed by G2 (196K), Forbes (181K), and Amazon (133K). This demonstrates a preference for established sources with structured data in the 2025 sample.

The practical implication is broader than "get on Wikipedia." Brands need accurate, consistent information on their own site and on independent sources buyers already use. Manufactured mentions are unlikely to help. Useful third-party coverage, reviews, research, and expert participation can.

Perplexity source preferences

Perplexity is more UGC-focused, with Reddit dominating citations (3.2M), followed by YouTube (906K) and LinkedIn (553K). This reflects Perplexity's semantic and vector-based approach to search in the 2025 sample.

For content creators looking to optimize for Perplexity specifically, focus on creating semantically rich content that thoroughly explores topics from multiple angles. The vector-based approach rewards content that demonstrates a full understanding of concepts rather than keyword-optimized writing that might perform well in traditional search engines.

That doesn't justify seeding promotional posts. It supports a more useful approach: answer real questions where your audience already discusses the topic, publish evidence on your own site, and make both easy to verify.

Google AI Overview source preferences

Google AI Overviews appears more domain-agnostic, with YouTube (406K), LinkedIn (384K), and Gartner (342K) leading citations. Reddit ranks fourth at 301K citations in the 2025 sample. Current Google guidance for AI features is clear: pages must be indexed and eligible to appear in Search with a snippet, and no extra AI-specific technical requirement is needed.

Microsoft Copilot source preferences

Copilot shows a strong preference for Forbes, with 2.1M citations, significantly higher than other platforms. Gartner follows at 1.3M citations in the 2025 sample. That concentration may reflect the commercial and informational prompts used, the retrieval system available at the time, and the authority of those sources for business queries.

Across the 1M+ prompts and keywords SEOmator tracks each month, the platform gap remains visible in a different way. Perplexity cited an average of 8.3 URLs per answer in our July 2026 snapshot, while ChatGPT search mode averaged 2.5.

What did the study say about traffic, backlinks, and content format?

One of the most surprising findings challenges fundamental SEO assumptions about what drives visibility. When analyzing the correlation between traditional SEO metrics and AI citation frequency, the researchers found that most established ranking signals had minimal direct correlation with AI search performance in the measured sample.

Traffic had a weak relationship with citations

The analysis revealed that 95% of AI citation behavior could not be explained by website traffic metrics (r² = 0.05). The data shows remarkable anomalies:

• Sites with effectively zero traffic can receive 900+ AI citations
• Sites with low traffic but high-quality, AI-friendly content often outperform high-traffic competitors
• High-traffic sites frequently receive disproportionately low citation counts

This weak correlation suggests that AI search is evaluating content through metrics separate from visitor popularity. Traffic is a poor stand-alone proxy for AI visibility. It doesn't mean traffic can never correlate with the authority, brand demand, or discovery signals that help a page enter the retrieval pool.

Backlinks also had a weak relationship with citations

Even more surprisingly, 97.2% of AI citations could not be explained by backlink profiles (r² = 0.038). The inverse relationship in the sample was particularly striking:

Similarly, 97.2% of AI citations cannot be explained by backlinks. Sites with fewer backlinks often receive significantly more AI citations than better-linked competitors.

This data suggests that AI search is evaluating content quality through metrics beyond traditional SEO signals like backlinks and traffic. The finding challenges the two-decade-old SEO principle that link building is the primary driver of search visibility.

Backlinks still affect traditional discovery, PageRank, authority, and rankings. The study tested a direct bivariate relationship with citation counts, not every indirect path through which links can influence retrieval. I read this as a measurement warning: don't use domain authority or backlink totals as your AI visibility KPI.

What did the presentation identify as citation factors?

If traffic and backlinks don't determine AI citations on their own, what does? The research indicates that the strongest factors influencing AI citation frequency are:

1. Content format alignment with AI preferences (comparative listicles, direct answers)
2. Semantic clarity in URLs, headings, and meta descriptions
3. Technical accessibility to AI crawlers (proper indexing and text availability)
4. Content recency, with newer content receiving preferential treatment in the sample
5. Structured presentation that facilitates easy information extraction

This decoupling from traditional ranking signals creates both a challenge and an opportunity. Organizations willing to recalibrate their content strategies for AI consumption patterns can achieve visibility regardless of their historical SEO performance or domain authority, but the list above should be treated as a set of testable hypotheses rather than proven causal factors.

Comparative content appeared most often

Analysis of 177 million sources cited in AI search results reveals clear patterns in content preference:

Comparative listicles dominate AI citations, accounting for nearly a third of all citations in this sample. This directly challenges conventional SEO wisdom that favors long-form, in-depth content. For AI search, well-structured comparative content appears to be substantially more valued.

Comparative content is often easy to retrieve because it names entities, applies repeated criteria, and states differences directly. Yet the table describes what was cited. It does not prove that changing an article into a listicle will cause more citations.

Which technical requirements still matter in 2026?

Technical accessibility remains the entry condition. The 2025 presentation was right to emphasize crawlability and text availability, but current platform documentation gives a more precise checklist than "AI crawlers don't interact with JavaScript."

However, being indexed is only the beginning. Many other technical factors determine whether your content will be cited by AI search engines.

Make important content available as text

JavaScript is not automatically an AI-search failure. The real risk is that the answer-bearing content is missing from the HTML or rendered output a crawler receives, hidden behind an interaction, or blocked by infrastructure.

When I review technical SEO guidance, I start with the output, not the framework. Can the crawler fetch the URL, receive a successful response, see the main text, follow its internal links, and identify the canonical page? A rendering comparison from an SEO crawler can find cases where raw and rendered content diverge.

Configure crawler access by purpose

OpenAI's publisher guidance separates OAI-SearchBot, which supports search visibility, from GPTBot, which is used for potential training. It also says ChatGPT referral URLs include utm_source=chatgpt.com.

Perplexity's robots.txt policy says PerplexityBot won't index the full or partial text of a disallowed site. Audit CDN and firewall rules as well as robots.txt; an allow rule in one layer can be blocked in another.

Treat llms.txt as optional

llms.txt is a community proposal for giving language models a curated Markdown map of a site at inference time. It can be useful for documentation discovery or controlled experiments. It is not a substitute for indexable pages, a sitemap, internal links, or crawler access.

Google explicitly says site owners do not need a new machine-readable AI file to appear in AI Overviews or AI Mode. If you want to test the proposal, first understand what llms.txt does, then keep the file factual and maintainable. SEOmator's LLMs.txt generator can create a starting version from a sitemap.

What content should you create for AI search?

Content still has to solve the query. The strongest transferable lesson from the study is not "write listicles." It is to make facts, comparisons, and decisions easy to verify and extract without stripping away the detail a human reader needs.

Prioritize these content properties:

• Direct answers: Answer the likely question near the start of the relevant section.
• Specific evidence: Name the sample, date, scope, and source beside each claim.
• Consistent entities: Use stable names for the brand, product, category, and alternatives.
• Real comparisons: Apply the same criteria to every option and show the result in a semantic table.
• Textual access: Keep important facts in visible text, even when a chart or video supports them.
• Original information: Publish data, workflows, or expert judgment that another page cannot cheaply reproduce.

I would not convert every article into a comparative listicle. I would use a comparison when the reader is making a choice and a guide when the reader needs a process. Format should follow the job.

Another critical content finding: AI search engines heavily favor recent content. The 2025 analysis showed that AI search engines could pick up and cite content on the scale of days, not weeks or months.

This recency bias creates both challenges and opportunities for SEO professionals. While it means content can become outdated quickly, it also means new content can gain visibility rapidly, a stark contrast to traditional SEO where ranking improvements often take months.

The finding does not promise that an update will earn visibility in days. Update a page when the facts, product, query set, or evidence changed. Don't change the date alone.

In specific contexts, user-generated content and social media can play a significant role in AI search results. For technical, rapidly evolving topics (like cloud GPU providers), AI search engines show a notable tendency to cite Reddit threads and other community content.

For more current benchmarks beyond the 2025 presentation, use our AI SEO statistics reference. It separates source dates and methodologies so you can compare studies without blending unlike query sets.

How should you implement an AI search optimization strategy?

Based on the research findings, here's a practical approach to optimizing for AI search. Start with measurement, then fix access and content gaps in that order. AI visibility can change between engines and model updates, so a repeatable query set is more useful than an isolated screenshot.

1. Track relevant queries

Monitor how AI systems respond to queries related to your brand, products, and industry. Unlike traditional search, AI search visibility can change rapidly, making regular monitoring essential.

Build a fixed set of prompts tied to your buyer journey:

• Unbranded discovery prompts, such as "best software for [use case]"
• Comparison prompts that name two or more options
• Problem-solving prompts around features, integrations, or constraints
• Branded prompts that test factual accuracy and positioning

Run each prompt across the engines your buyers use. Record whether your brand is mentioned, whether your domain is cited, which page earns the citation, and how the answer describes you. A GEO audit can establish the baseline across repeatable prompts.

2. Create AI-conducive content

Develop content that aligns with AI citation preferences:

• Format: Prioritize comparative listicles, well-structured Q&A content, and formats that directly answer specific queries
• Recency: Update content when its facts, evidence, or product details change
• Semantic structure: Use clear headings, structured data, and organized information that AI can easily parse
• Direct answers: Provide clear, concise answers to likely queries directly in your content

3. Monitor and mimic citations

Study which of your competitors' pages receive AI citations and analyze their structure, format, and content approach. Adapt your content to incorporate successful elements while maintaining originality. Citation tracking can help identify which competitor pages are gaining traction in AI search.

4. Check technical eligibility

For every page you want cited:

1. Confirm it returns a stable 200 response.
2. Check canonical and indexation signals.
3. Verify the main answer is visible in fetched or rendered text.
4. Confirm internal links make the page discoverable.
5. Review search-crawler and user-fetch crawler rules.
6. Check whether the CDN or firewall blocks published bot IP ranges.

5. Compare the cited pages

Study the pages that already receive citations for your prompt set. Compare their evidence, page type, source reputation, freshness, and answer passage. Don't mimic wording. Identify the missing fact or perspective that your page can own.

6. Improve one query cluster at a time

Revise the page around the query's actual decision:

• Add a concise answer where the current page delays it.
• Replace unsupported claims with sourced facts.
• Turn a genuine multi-option comparison into a table.
• Add original data or a documented workflow.
• Correct inconsistent product and entity descriptions.

7. Build independent corroboration

The AI system may rely on sources beyond your site. Keep product facts consistent across credible review pages, industry publications, expert profiles, videos, and community discussions. Earn those mentions by contributing useful evidence, not by manufacturing praise.

8. Re-run and log the result

Compare mention rate, citation rate, cited pages, and answer context against the baseline. Keep the prompt wording and location settings stable enough to make the comparison meaningful. Model variance remains, so use repeated runs rather than a single pass.

How should you measure AI search performance?

Track metrics specific to AI search. Citation presence tells you whether an engine used your source. Referral sessions and conversions tell you whether that exposure produced measurable visits:

• Citation frequency: How often your domains/pages are cited in AI responses
• Citation share: Your percentage of citations compared to competitors for key queries
• Citation context: How your brand/content is characterized in AI responses
• Visibility score: Aggregate metric tracking overall AI search performance

Google announced generative AI performance reports in Search Console on June 3, 2026. The initial rollout to a subset of sites includes impressions, pages, countries, devices, and time trends for generative AI features in Search and its content feed.

ChatGPT referral links can be segmented with the utm_source=chatgpt.com value documented by OpenAI. For visibility without a click, use prompt tracking. SEOmator's AI brand visibility checker, AI Overview keyword checker, and Google AI Mode checker cover different parts of that measurement problem.

Our July 2026 data also shows why branded tests are insufficient. Across the 1M+ prompts and keywords we track monthly, branded-prompt mention rates were above 91% across the measured engines, while discovery-prompt rates fell to 11%–18%. The contest is whether an engine names you before the user supplies the brand.

What are the study's limits?

The presentation is a large descriptive study, but size does not remove sampling limits. Query intent, industry mix, geography, model version, retrieval configuration, and collection date can all affect the source set.

Use the findings with four constraints:

1. Separate the experiments. The 41M corpus, 650-execution overlap test, 50K-prompt correlation analysis, and 177M-source content classification answer different questions.
2. Don't infer causation from correlation. Low r² values show poor stand-alone explanatory power, not that the signals have no indirect effect.
3. Treat platform preferences as dated snapshots. A source mix observed in April 2025 may change after a model, index, or interface update.
4. Validate on your own prompts. A B2B software query set may behave differently from local, health, news, or e-commerce queries.

SEOmator's first-party figures in this update come from a separate B2B-heavy panel: about 60% B2B SaaS, 20% e-commerce, and 20% mixed businesses. The July data covers July 1–21, 2026. We report it as a snapshot, not a completed monthly trend.

Frequently asked questions about AI search optimization

What is AI search optimization?

AI Search Optimization is the practice of optimizing digital content to increase visibility and citation frequency in AI-powered search systems like ChatGPT, Google AI Overviews, Perplexity, and Microsoft Copilot. It combines elements of traditional SEO with new techniques specific to how AI systems process and surface information.

How is AI search optimization different from traditional SEO?

Traditional SEO focuses on ranking websites in search engine results pages (SERPs), while AI Search Optimization focuses on getting your content cited within AI-generated answers. The key differences include the importance of different ranking signals (backlinks matter less), content format preferences (listicles perform better), and the direct relationship between users and AI rather than users and websites.

Those parenthetical claims describe the presentation's 2025 sample, not universal rules. The same crawlability and content-quality foundations support both disciplines.

How do I get my website cited by ChatGPT?

Make the target page public, crawlable, text-accessible, and useful for the query. Allow OAI-SearchBot through robots.txt, hosting, and CDN controls if you want the content included in ChatGPT summaries and snippets. Then test a fixed set of prompts and compare your page with the sources ChatGPT already cites.

How do I appear in Google AI Overviews?

Google says a page must be indexed and eligible to appear in Search with a snippet. There is no extra AI-specific markup or file requirement. Apply standard Search guidance: allow crawling, use clear internal links, keep important content in text, provide a good page experience, and ensure structured data matches visible content.

Does llms.txt improve AI search rankings?

No official source cited in this article identifies llms.txt as a ranking requirement. It is a community proposal for pointing language models and agents to curated Markdown resources. Test it if your documentation use case fits, but prioritize crawl access, indexation, sitemaps, internal links, and accurate page content.

Do backlinks matter for AI citations?

No. The research demonstrates that traditional ranking signals like backlinks and domain authority have minimal correlation with AI citation frequency in the measured sample. Of AI citation behavior, 95% could not be explained by traffic metrics, and 97.2% could not be explained by backlink profiles.

That direct correlation result does not make links irrelevant. Links can still support discovery, traditional rankings, authority, and brand visibility. Track citations directly instead of treating backlink totals as a substitute for AI visibility.

Which content formats perform best in AI search?

Comparative listicles dominate AI citations, representing 32.5% of all citations across platforms in the 2025 dataset. Other high-performing formats include opinion blogs (9.91%) and detailed product or service descriptions (4.73%).

Use that result as a descriptive pattern. Choose a comparison format when the query requires evaluating options; use a guide, definition, or data report when that better serves the reader.

How quickly can AI search optimization show results?

There is no reliable universal timeline. The presentation observed that new content could be cited within days, but crawl timing, indexation, query demand, source competition, and model updates vary. Establish a baseline, make one substantive change, and repeat the same prompt set over several runs.

Can I optimize for every AI search engine with the same approach?

While core principles apply across platforms, each AI search engine shows distinct preferences. ChatGPT favored Wikipedia and established reference sources, Perplexity prioritized UGC content like Reddit, Google AI Overviews appeared domain-agnostic, and Microsoft Copilot heavily favored Forbes and other business publications in the 2025 sample.

The foundations transfer: crawlable pages, useful text, evidence, consistent entities, and independent authority. The source set and citation capacity vary by engine. Track each target platform separately and prioritize the two or three that your buyers actually use.

What tools can help me track AI search performance?

Specialized tools from companies like Profound, BrightEdge, and Semrush have emerged to track AI search visibility. These tools monitor citation frequency, citation share relative to competitors, and analyze how AI systems characterize your brand and content.

SEOmator's tools add query-level checks for AI brand visibility, Google AI Overviews, and Google AI Mode, along with repeatable GEO audits.

What should you do next?

AI search has created a second visibility layer on top of traditional search. The 2025 study showed that citations can diverge from Google and Bing results, that traffic and backlink totals are weak proxies for citation frequency, and that each platform assembles a different source mix.

The practical response is measured, not speculative. Make important pages crawlable and indexable. Publish facts worth citing. Track mentions and citations for a stable set of discovery prompts. Then improve the page or off-site evidence where the citation gap is visible.

Start with a stable query set and record which target queries trigger an AI Overview, which brands the major answer engines cite, and which source pages support their claims. That gives you a baseline you can test instead of a list of unproven AI-search hacks.

As Blyskal noted in his presentation: "As former SEOs, we're the black sheep of marketing. AI search is about to become the sexiest area of digital marketing."