AI SEO Strategy: What Gets a Brand Named by AI, Tested on 100 Buying Questions

What This Study Says About Your AI SEO Strategy

My AI SEO strategy used to rest on other people's numbers. This page replaces that with my own test, and one of my old claims came out much weaker than I wrote it.

Here is my read of where to spend your time, based on this data, in order.

• First, earn a spot on the lists that rank. Brands featured on a publisher or review-site page in Google's top 10 were named 5.5 times as often, and the lift held for challenger brands too (6.1x).
• Second, keep your own site ranking. It helps too, at 2.3 times as often, a smaller lift than the list signal in this sample.
• Don't treat Google rankings as the gate. About two in three citations pointed to sites outside Google's top 20 for the same search.
• Plan for ChatGPT separately. It linked to the brand it named first in 75.0% of answers, while Claude and Perplexity mostly linked to third-party lists.

What Gets a Brand Named by AI

This is the question most AI SEO advice skips. For the 37 open-ended buying questions with a verified Google page, Claude agents checked every brand in the category against two things: was it featured on a publisher or review-site page in Google's top 10, and does its own site rank in the top 20?

Then they checked whether each AI engine named it. A brand on one of those pages was named in 34.1% of answers, against 5.0% for brands on none.

Popularity is the obvious objection. HubSpot shows up on CRM lists and in CRM answers because it's HubSpot.

So the analysis split each category's brands into thirds by Google search volume and compared brands within the same third. The list signal held at 5.5 times, and for challenger brands outside each question's three leaders it was 6.1 times (95% range 4.6 to 9.9).

It also held for brands whose own site wasn't in Google's top 20 (5.3 times, range 3.9 to 8.5). For a smaller brand, that is the most useful number on this page.

Do You Need to Rank on Google to Get Cited by AI?

That old claim, that a site that can't rank generally won't get cited, isn't supported by this sample.

Only 35.7% of citations pointed to a site in Google's top 20 for the same search, far short of the 60% my plan set to confirm the claim, though not low enough to formally reject it. At the top 10 it was 26.5%.

ChatGPT was furthest from Google at the page level. Just 8.9% of the pages it cited were in Google's top 20 for the search, against 26.0% for Claude.

Outside research found a similar pattern, though measured differently. Ahrefs found that only 12% of links cited by ChatGPT, Gemini and Copilot appear in Google's top 10.

One complication is worth knowing about. On 28 of the 60 keywords that could be checked, the live Google page the tools pulled was poisoned, with the usual sites swapped out.

Those keywords were measured against a recent clean snapshot instead. The AI Overviews made the problem plain: on the 26 AI Overviews that loaded for the 32 keywords whose live page checked out clean, their citations matched it 37.2% of the time, against 5.5% on a poisoned page.

What Gets Cited: Third-Party Lists

73.0% of citations went to third-party sites rather than to the brands themselves. That held even with retailers like Amazon counted on the brand side (67.9%).

And the third-party pages that get cited are mostly lists. On best, budget, how-to-choose and problem questions, best-of lists made up 69.5% of third-party citations.

This is why I think the lists matter so much. On those questions, when an AI cited a third-party page it was usually the same kind of page you'd read: someone else's ranked list.

ChatGPT vs Claude vs Perplexity vs Google AI Overviews

The engines didn't behave the same, and two differences change what you should do.

ChatGPT linked to vendors far more than the rest. 48.0% of its citations went to brand sites, against 20.9% for Claude and 22.2% for Perplexity.

Google AI Overviews leaned on video. YouTube made up 27.3% of its citations (208 of 762), while ChatGPT, Claude and Perplexity didn't cite YouTube once in this sample.

Search itself is not a given either. In a separate run where Claude could choose, it searched the web on 68 of the 100 questions and answered the rest without searching, and Gemini searched on 81% on the first try.

My read: an answer without search leans on what the model learned in training. Training data also comes from what other sites wrote about you, though this study didn't test that link.

Explore a Category

Pick a category and an AI engine to see which brands it named most and which sites it cited most across that category's 10 questions. Brands named in the question itself aren't counted for that answer.

The AI SEO Strategy I'd Run Now

The five moves below are the process I use, updated for what the data showed. They run in sequence, and move four is where this study found the strongest link.

1. Pick the one thing you want to be known for

Before any tactic, write a positioning line short enough that a model can repeat it without garbling it. Mine is "AI-first marketer."

This study didn't test positioning, so this part is my own practice. My reasoning: if your site, socials and podcast bios describe you three different ways, a model may not have a confident answer, and it can name someone else.

2. Say the same thing everywhere a crawler can reach

Repeat the line on your site, every social bio, review profiles like G2 and Trustpilot, and directory listings. Then open your robots.txt and confirm you aren't blocking OAI-SearchBot, Claude-SearchBot or PerplexityBot.

My free agent readiness checker shows which of them your homepage blocks in about 10 seconds. Those bots also read different search indexes, which may help explain why each engine cited such different pages here.

One correction plenty of guides get wrong: Google-Extended controls whether your content is used to train Gemini and ground its answers, per Google's crawler documentation. Google says it doesn't affect inclusion in Google Search, and its AI features guide points to it only for Google's other systems, so it isn't an AI Overviews switch.

3. Publish pages built to be quoted

Your own best-of, comparison and alternatives pages are still worth writing. Just know who reads them: ChatGPT cited brand sites heavily, Claude and Perplexity mostly cited third-party lists, and AI Overviews leaned on YouTube.

Write each recommendation with its condition in the same sentence, like "Tool X wins if you need Y." AI answers often quote a single passage, so a passage should never travel without its condition.

Question-style headings, a direct answer in the first two sentences, specific numbers and self-contained table rows do most of the formatting work. For an example, Sam Dunning walked me through an alternatives page that ranked and got cited.

That checklist is the part worth automating, since a person skips steps on deadline and a skill file runs the same steps every time. I run it as Claude Code skills, and I broke down the ones I use in my guide to Claude Skills for SEO.

4. Get on the lists that already rank

This is the move with the strongest link in the data. Here's the workflow I'd run for each buying question in your category.

• Pull the Google top 10. Search the question the way a buyer would, like "best CRM for small business."
• Keep the publisher and review pages. Skip vendor pages, forums and stores, and note every publisher or review page that features brands.
• Check who's on them. If your competitors are featured and you aren't, that list is your target.
• Earn the spot. Pitch the author with a specific reason you belong, offer a trial or review unit, or join the publisher's affiliate program if it has one.
• Add video for Google. AI Overviews cited YouTube heavily, so a clear review or comparison video is worth making for your top questions.

Podcasts and digital PR still help, since they create pages about you on other people's sites. I run the PR side through an automated pitching tool, and I show the real numbers in my Backlinker AI review.

5. Measure whether AI names you

Write down 10 questions a buyer would ask and run them monthly in ChatGPT, Claude, Perplexity and Google. Log whether your brand gets named, and which lists get cited when it doesn't.

Judge the channel on signups, since sessions understate its value. Ahrefs reported that 12.1% of its signups came from AI search, from a far smaller share of its traffic.

For the tracking side, I broke down how to track AI visibility and what paid trackers miss in my Searchable review.

Do you need an AEO retainer?

My take: usually not. Stackmatix puts small-business AEO retainers at $1,500 to $5,000 a month, and Digital Elevator's guide has enterprise programs at $10,000 to $30,000 and up.

Most of that buys schema, content structure and crawlability. The piece worth paying for is outreach, the expert quotes and pitches that get you onto the lists that rank, so ask any provider which lists they'll get you on and how they'll check that AI names you afterward.

If you're comparing help, see who is worth paying for AI search work and what each price tier buys. My own AI SEO System packages this workflow, so weigh my bias.

How This Compares to Other AI Search Studies

Other studies measured pieces of this. As far as I can find, none joined Google rankings, citations and brand mentions on the same questions across all five engines and published the rows.

• Google rank overlap. Ahrefs found that 37.1% of URLs cited in AI Overviews ranked in Google's organic top 10, against 12% for ChatGPT, Gemini and Copilot.
• Cited vs named. Semrush found that only 6-27% of the most mentioned brands also rank as top cited sources. Its ghost citations study logged 3,981 domain appearances and found 61.7% were cited as a source while the brand never appeared in the answer.
• Brand sites vs third parties. Yext found that 80% of citations pointed to a brand website or a listing. That study asked location-aware brand questions and counted directory listings as brand-controlled, so it measures a different question than these category buying questions.

How I Ran This Study

My plan, metrics and pass-fail zones were written down, reviewed twice and frozen before any result was computed. A small pilot had already been pulled by then, and the check for poisoned Google pages was added after the first pull showed one.

• Questions. 10 per category, from the same template: three "best" questions, two head-to-heads, two "alternatives to" questions, "how do I choose", "is it worth it" and one problem described in plain words. The brands in each question were three category leaders, ordered by Google search volume.
• Engines. ChatGPT through the chatgpt.com answer with web search forced on, and Google AI Overviews from the live results page for the matching keyword. Both were collected October 22, 2026, from the US, with DataForSEO.
• API engines. Claude (claude-sonnet-5, search forced on), Gemini (gemini-3.5-flash, served as gemini-3.6-flash) and Perplexity (sonar) ran through their APIs with web search on. API answers can differ from the consumer apps.
• Who did the work. Claude agents collected, labeled and analyzed the data. They labeled every cited site by owner (brand site, publisher, review platform, forum, video, retailer or reference) and every cited page by type.
• Brand detection. The agents built a brand list per category, 531 brands in total. It has rules for names that double as everyday words, like Ring, Purple and Shark.
• Checks. A separate Claude model re-checked fixed random samples without seeing the original labels. It agreed on 48 of 50 site labels, 47 of 50 page types and 50 of 50 brand detections.
• Stability. A rerun of 20 questions on four engines (15 for Gemini, which skipped search on some) brought back 78% to 94% of the brands named the first time.
• Pooling and coverage. Each unique URL cited in an answer counts once, and pooled figures weight each engine equally. AI Overviews loaded on 87 of the 100 keywords.
• Google reference. Rank trackers are getting poisoned results pages, so each keyword's live pull was checked against DataForSEO's recent snapshots. A pull missing most of the keyword's usual sites was set aside for the newest clean snapshot.
• Keywords left out. 40 of 100 keywords lacked enough history to check, so they were left out of the rank figures for ChatGPT, Claude, Perplexity and Gemini. AI Overviews were compared with their own live results page, with poisoned ones reported separately.
• The search-volume adjustment. Brands were split into thirds by Google search volume within each category, and the 5.5x and 2.3x figures compare brands within the same third and engine (a Mantel-Haenszel ratio). 95% ranges come from resampling questions 2,000 times.
• Gemini. It skipped web search on some questions even after two reruns, so it searched on 86% of questions, under the 90% bar I set. Its numbers are shown but left out of every pooled figure.
• Pages that couldn't be read. 73 of 370 ranking pages couldn't be read by any fetch method the agents tried. 19 of those were publisher or review pages (10 behind a bot check), from forbes.com, g2.com, gartner.com, getapp.com, moderncastle.com, sleepfoundation.org, solutionsreview.com, thedigitalprojectmanager.com, and they count as not featuring any brand.
• Disclosure. I sell Claude Code products, an AI SEO system and consulting, and Claude is one of the engines tested and did the labeling. No category here includes my business or my clients'.

The limits matter. The analysis uses one retained answer per question and engine, collected on one day, so the figures describe this sample, and answers shift from run to run.

The list signal is a strong association, and it doesn't prove cause. A brand can land on lists and in AI answers for the same underlying reason, even after the search-volume adjustment, and the list and own-site signals weren't adjusted for each other.

My plan also tested whether the first brand an answer names usually goes uncited on its own site. Pooled, the first-named brand was cited on its own site in 31.9% of answers, but the 95% range (25.6% to 38.1%) crosses my 35% line, so I report the engine split above and draw no pooled conclusion.

Ryan's Final Thoughts

I expected the data to confirm that ranking on Google is the foundation of AI visibility. In this sample it showed something more useful: being featured on the publisher and review pages that rank was the stronger of the two signals measured.

So I'd spend the next quarter getting onto those lists, one buying question at a time. Then I'd check every month whether the answers changed.

Doser, R. (2026). AI SEO strategy: what gets a brand named by AI, tested on 100 buying questions. ryandoser.com. https://ryandoser.com/ai-seo-strategy/