How AI Recommends Running Brands in 2026: The Largest Study Yet.

We ran 2,200 unaided prompts across ChatGPT, Claude, Gemini, Perplexity, and Grok to see which running and endurance brands AI recommends, and why nine brands own the conversation.

A growing share of runners have stopped Googling “best trail shoe” and started asking ChatGPT, Gemini, or Claude. Whatever the model says back is the new top of the funnel. I wrote about that shift a while ago in The New Top of Funnel Is a Chat Reply.

The New Top of the Funnel Is a Chat Reply

This is the follow-up, with receipts.

Working with Michael Rueckert of Centium AI, we ran the largest study Centium has done (as of June 2026, we will do this semi-annually!): 22 categories, 20 prompts each, across the five biggest models (ChatGPT, Claude, Gemini, Perplexity, and Grok), for 2,200 total responses. Every prompt was unaided. We asked the category question (”what are the best running shoes,” “best electrolyte mix,” “best running town”) and never seeded a brand name. We ran them through the API, with no chat history and no location, so the model had nothing to anchor to. Whatever it recommended, it recommended on its own.

• 2,200 responses

• 22 categories

• 2,052 brands and entities

• 16,990 citations

• 2,558 domains

The headline is not what small brands would be excited to hear: AI is not the level playing field the technology is assumed to produce. It has favorites, it formed them in public, and for the first time you can see exactly how.

This write-up is a companion piece to a 90 minute podcast with Michael Rueckert, Founder of Centium AI. The full episode can be found on the Long Run Labs Podcast on Apple and Spotify, and you can watch it on YouTube too!

Key findings (the short version)

• Nine brands account for 22 percent of all AI mentions in running, while 56 percent of the brands that surfaced appeared exactly once and never again.

• Authority does not transfer across categories. Winning “best running shoe” does not win “best running headphones.” You earn each category separately.

• AI rewards fresh sources. The median cited source is 4.1 months old once weighted by use, and 48 percent of everything cited was published in the last six months.

• “Best of” and “top” roundups are 53 percent of all citations. The ranked list is the unit of recommendation.

• The average AI answer names just 6.5 brands. You are one of about six names, or you are invisible.

What is inside

• I tested this on myself first

• How AI actually decides what to recommend

• Nine brands own the conversation

• Winning one category does not win the next

• Where the recommendations actually come from

• What this means for you

• The uncomfortable part

• The bottom line

• FAQ

I tested this on myself first

Before I trust a finding, I like to break it. So on a Sunday, I built one page on my site: a ranked “best running and endurance podcasts of 2026.” Nothing fancy. A list with commentary, aligned with the insights from the research on how my own podcast is showing up in the “eyes” of the LLMs.

By Thursday it was the single most-visited page on my entire site all time, not counting pages linked from the header. It ran about 2x my homepage in the week. A page four days old beat everything I had published in months. It is still the most popular page, and growing fast.

That is the whole study in one example. The models are out there searching, they are hunting for a very specific kind of content, and when you hand it to them, you show up. Everything below explains why that worked, and shows you exactly who is winning what.

How AI actually decides what to recommend

Michael uses a useful framework for this. Picture the model taking an exam.

The closed-book part is its training data, the instinct in its head. It has read enormous amounts about Nike, HOKA, and Adidas, so when you ask a running question, those names surface first by reflex. That is why the biggest incumbents own the broad questions.

The open-book part is what happens next. The model goes and searches the live web, a behavior that only really switched on across these tools in early 2025. It breaks your question into its own micro-searches, a process called query fan-out, and pulls in what it finds. Here is the key: because the models were trained on the old internet, when they reach for the live web they are not looking for what they already know. They are looking for what is new.

You can measure that bias precisely. We indexed the publish date of the top 250 sources by citation volume and recovered dates for 219 of them. The median cited source is 6.5 months old, and once you weight by how often each source actually gets used, that drops to 4.1 months. Nearly half of everything cited, 48 percent, was published in the last six months. 39 percent came out in the last three. The sources the models lean on hardest are the freshest ones.

And you can see what they search for. Across 8,799 follow-up searches, the behavior is pure informed shopper:

What the models put in their own searches (share of searches)

• Include a year (19% say 2026, 18% say 2025): 40%

• Include “best”: 37%

• Ask for a “review”: 24%

• Include “top”: 12%

• Qualify for a person or use case (for women, for beginners, for flat feet): ~5%

• Direct “X versus Y” comparison: 2.3%

Put the format words together and roughly three-quarters of the models’ own searches are hunting for a best, a top, or a review. They search “best X 2026,” they find a ranked list, and they recommend whoever is on it. That is why my Sunday page worked. I had written the exact artifact the machines were out looking for.

And you can watch them do it. I pulled my crawler logs, which are just the record of every automated bot that visits a site, and across my whole site, AI crawlers now account for roughly half of all bot traffic. The most active crawler on my entire site is not Google, it is Meta’s AI agent, with ByteDance’s right behind it, and Anthropic’s ClaudeBot, OpenAI’s GPTBot, and OpenAI’s search bot all working the site constantly. These are the exact systems this study measured, and I can watch them reading.

You can even see both halves of the “exam” in the log. The training crawlers are quietly building the closed-book instinct for the next generation of models. The retrieval bots, the ones that fetch a page at answer-time, are the open-book fan-out happening live. Both are on my site right now, and I am blocking none of them (some people do this because they want the site traffic over AI crawlers). That is the point. If invisible to AI means invisible, the last thing you do is close the door.

(If you want the genuinely good, no-math version of how these models work under the hood, Andrej Karpathy’s Deep Dive into LLMs like ChatGPT is the one to watch. Karpathy was a founding member of OpenAI and led AI at Tesla, and he is the rare researcher who can explain next-token prediction to a normal human.)

Nine brands own the conversation

Now the concentration, and it is steep. Of the 2,052 brands and entities that surfaced, only nine were named in at least 10 percent of all prompts, and just 126 cleared even 1 percent. At the other end, 1,149 of them, 56 percent of everything that came up, appeared exactly once across all 2,200 responses and never again.

The top nine brands account for 22 percent of all mentions. The top 25 for 35 percent. The top 100 for 61 percent. And the average answer named just 6.5 brands while citing 7.7 webpages. There is no page two in an AI answer. You are one of about six names, or you are invisible.

Here is the overall board, every category blended together (mention rate):

All prompts (overall)

1. Nike (23%)

2. Brooks (20%)

3. HOKA (19%)

4. Adidas (16%)

5. Saucony (16%)

6. Salomon (15%)

7. New Balance (15%)

8. ASICS (14%)

9. On (10%)

10. Altra (8%)

11. Garmin (8%)

12. lululemon (8%)

13. PUMA (7%)

14. Nathan (7%)

15. Patagonia (6%)

If you are a smaller brand, this reads like a rich-get-richer machine, and at the broad level it partly is. The model’s instinct was built on a decade of the internet talking about the giants. You cannot out-instinct that overnight. But the broad question is not where most runners actually shop, and the moment you go category by category, the whole picture cracks open.

Winning one category does not win the next

This is the most important finding in the study, and the most hopeful one. Ask any model for “the best running brand” and the same incumbents return (mention rate):

Running brands (the broad question)

1. ASICS (88%)

2. HOKA (88%)

3. Nike (87%)

4. Brooks (85%)

5. Saucony (82%)

6. Adidas (79%)

7. New Balance (75%)

8. On (62%)

9. Mizuno (44%)

10. Altra (37%)

Now watch that grip dissolve. Move category by category and a different brand leads almost every one. Several brands dominate a niche yet score a flat zero in the broad question: Beats (67 percent of headphones) and Bose (58 percent), zero in “best running brand.” Science in Sport (66 percent of gels), zero. TriggerPoint (26 percent of recovery), zero. Authority does not transfer. Owning “best running shoe” and owning “best running headphones” are two unrelated wins, and you earn each one separately.

The reason is mechanical. The model cannot try on the shoes, taste the gels, or feel the watch, so for each category it goes to that category’s specialists. Ask about headphones and it does not lean on Runner’s World, it goes to tech and ratings sites (per Centium’s dashboard, rtings.com was cited in 81 percent of headphone answers). Ask about gels and it goes somewhere else entirely.

The takeaway for any brand that is not Nike: you do not have to beat Nike. You have to own your category. Here is the full board, category by category.

Shoes

Road running shoes

1. Brooks (69%)

2. ASICS (68%)

3. HOKA (68%)

4. Nike (62%)

5. New Balance (57%)

6. Saucony (56%)

7. Adidas (45%)

8. Mizuno (25%)

9. PUMA (25%)

10. On (23%)

Carbon plate racing

1. Nike (79%)

2. Adidas (78%)

3. ASICS (74%)

4. PUMA (61%)

5. New Balance (59%)

6. Saucony (57%)

7. HOKA (54%)

8. Brooks (30%)

9. On (20%)

10. Mizuno (19%)

Trail running shoes

1. HOKA (73%)

2. La Sportiva (67%)

3. Salomon (66%)

4. Saucony (61%)

5. Altra (57%)

6. Brooks (47%)

7. Adidas (35%)

8. Nike (30%)

9. New Balance (29%)

10. Topo Athletic (25%)

Gravel running

1. HOKA (66%)

2. Salomon (62%)

3. Altra (53%)

4. Nike (51%)

5. Brooks (50%)

6. Saucony (43%)

7. Craft (31%)

8. New Balance (27%)

9. Merrell (22%)

10. On (22%)

Note how completely the leaderboard reshuffles, and who benefits. HOKA is the standout, leading three of the four shoe categories (trail, gravel, and a three-way tie at the top of road) while sitting only third in the broad question. Brooks tops road outright. The trail and gravel categories belong to the specialists: La Sportiva is second in trail and absent from road entirely, the purest “own your niche” result in the study. Salomon is everywhere it should be, second in gravel and third in trail, and as you will see it owns the hydration categories too, making it arguably the most complete trail-and-mountain brand in the data. Altra has quietly built real cross-category presence, fifth in trail and third in gravel, and is one of only ten brands to crack the broad top ten. And Mount to Coast, a genuinely young brand, has already broken into a category (twelfth in gravel at 17 percent), which is exactly the playbook this whole piece argues for: pick one lane and claim it. Kiprun shows up only in road (fourteenth, 3 percent). Picking your category is the whole strategy.

Watches, apparel, and bras

Running watches and GPS

1. COROS (91%)

2. Garmin (91%)

3. Apple (65%)

4. Suunto (65%)

5. Polar (53%)

6. Amazfit (30%)

7. Google Fit (5%)

8. Huawei (4%)

Running apparel

1. Nike (60%)

2. Patagonia (57%)

3. Brooks (55%)

4. lululemon (53%)

5. Tracksmith (53%)

6. Adidas (52%)

7. Janji (41%)

8. On (35%)

Sports bras

1. Brooks (70%)

2. lululemon (69%)

3. Nike (46%)

4. Janji (39%)

5. Panache (39%)

6. Shefit (37%)

7. Oiselle (36%)

8. rabbit (29%)

Watches is the most top-heavy category in the entire study: Garmin and COROS are tied at 91 percent, and the next brand (Apple) is a cliff away at 65 percent. If you are not one of those two, you are barely in the conversation.

Two things worth pulling out of the apparel and bra boards. Brooks shows up again, third in apparel and first in sports bras at 70 percent, which makes it one of the genuinely rare multi-category winners in the whole study: number one in road shoes, number one in sports bras, top three in apparel. Most brands own one lane. Brooks owns three. And On is the interesting counter-pattern. It does not lead a single category here, but it appears across road, gravel, and apparel and is one of only nine brands to clear 10 percent overall. That is a different and real kind of strength, broad familiarity rather than category ownership, and it is the inverse of the specialist’s path.

Hydration and carry

Hydration vests and packs

1. Salomon (87%)

2. Nathan (82%)

3. Ultimate Direction (77%)

4. UltrAspire (56%)

5. CamelBak (52%)

6. Osprey (46%)

7. The North Face (38%)

8. Patagonia (36%)

Water bottles and soft flasks

1. HydraPak (71%)

2. Nathan (59%)

3. Salomon (49%)

4. CamelBak (46%)

5. Osprey (37%)

6. Amphipod (28%)

7. Ultimate Direction (24%)

8. Hydro Flask (17%)

HydraPak topping water bottles is a lesson on specificity. We split bottles out as their own category on purpose. Before that, the brand was an OEM footnote inside the hydration vest answers. Carve out the category and a new leader appeared, which is a reminder that AI models have different favorites for every segment of the industry.

Salomon deserves its own line here. It does not just win trail and gravel shoes, it leads hydration vests at 87 percent and sits third in bottles (and those bottles are usually made by HydraPak!), the kind of stacked, adjacent ownership that is hard to dislodge once the models have it. That is what owning a set of related categories looks like in practice.

Fuel

Energy gels and chews

1. Science in Sport (66%)

2. Maurten (61%)

3. GU (60%)

4. Honey Stinger (57%)

5. Hüma (46%)

6. Precision Fuel & Hydration (35%)

7. Clif (32%)

8. TORQ (28%)

Electrolyte and hydration mixes

1. Nuun (74%)

2. Skratch Labs (74%)

3. LMNT (66%)

4. Tailwind (57%)

5. Liquid I.V. (50%)

6. Gatorade (31%)

7. Maurten (30%)

8. Ultima (29%)

Supplements and creatine

1. Thorne (55%)

2. Optimum Nutrition (34%)

3. Momentous (31%)

4. Transparent Labs (24%)

5. Legion (17%)

6. GU (15%)

7. Klean Athlete (15%)

8. Maurten (15%)

Watch the cross-category leakage, or lack of it. Maurten is strong in both gels (second) and electrolytes (seventh). Science in Sport owns gels and barely registers in electrolytes (fourteenth, 11 percent). Precision Fuel & Hydration sits sixth in gels and twelfth in electrolytes (21 percent) despite “hydration” being in the name, because the models are not matching keywords, they are reading roundups. And Skratch Labs is the clean win here, tied with Nuun at the very top of electrolytes at 74 percent, a smaller brand that has clearly out-earned its size in the one category it cares most about. That is exactly what strong, sustained PR in a category looks like once the models start reading. Think about Nuun and how many ambassadors they have. They have had dozens (hundreds?) of people posting and writing about their products, every year for over a decade.

Tech and recovery

Running headphones

1. Shokz (83%)

2. Beats (67%)

3. Bose (58%)

4. Apple (51%)

5. JBL (36%)

6. Soundcore (36%)

7. JLab (34%)

8. Jabra (30%)

Recovery devices and tools

1. Hyperice (72%)

2. Therabody (71%)

3. TriggerPoint (26%)

4. Ekrin Athletics (25%)

5. Roll Recovery (16%)

6. Bob and Brad (15%)

7. HOKA (13%)

8. Air Relax (12%)

Training apps and coaching

1. Strava (67%)

2. Nike Run Club (63%)

3. Runna (59%)

4. Garmin (51%)

5. Runkeeper (49%)

6. MapMyRun (23%)

7. TrainingPeaks (20%)

8. Adidas Running (19%)

Strava beating every dedicated coaching platform is one of the more telling outliers. The models reach for the app people actually talk about, not the one with the most specialized feature set. And Runna at third is the exception that proves the PR rule: it is the most-cited brand-owned site in the whole study, because it publishes its own “best of” guides instead of just product pages.

The running experience

Races and events

1. Chicago Marathon (16%)

2. UTMB (15%)

3. Abbott World Marathon Majors (13%)

4. Berlin Marathon (11%)

5. Boston Marathon (11%)

6. Hardrock 100 (10%)

7. New York City Marathon (10%)

8. Western States 100 (10%)

9. Honolulu Marathon (9%)

10. Leadville Trail 100 (9%)

Running destinations and towns

1. Boulder, CO (55%)

2. Boston, MA (43%)

3. San Francisco, CA (41%)

4. New York City (40%)

5. Flagstaff, AZ (35%)

6. Chicago, IL (30%)

7. Portland, OR (29%)

8. San Diego, CA (28%)

9. London, UK (27%)

10. Chamonix, France (26%)

Races is the most wide-open category in the entire study. The leader sits at just 16 percent and 165 different events surfaced. Nobody owns “best races,” which is rare white space. Destinations is the opposite story: Boulder ran away with it at 55 percent, off a specific repeated narrative the models latched onto (200 miles of trails, live-high-train-low altitude, a genuine running culture, sourced largely from a Runner’s World “best cities” list and an Outside “best trail towns” list).

Runner-friendly hotels and travel

1. Westin (46%)

2. Hilton (41%)

3. Marriott (38%)

4. Hyatt (37%)

5. IHG (35%)

6. Fairmont (32%)

7. The Ritz-Carlton (26%)

8. Four Seasons (25%)

Running specialty retailers

1. Road Runner Sports (68%)

2. Fleet Feet (61%)

3. Running Warehouse (50%)

4. REI Co-op (37%)

5. DICK’S (34%)

6. Amazon (22%)

7. Zappos (19%)

8. Nike (18%)

Westin owning hotels at 46 percent is a clinic in claiming open space cheaply. It built a running concierge that knows the local routes, put it on its site, and the models latched on. No other chain was really trying to own “runner-friendly,” so Westin won it. And note that retailers surface most when the question carries buying intent (”where can I buy”) rather than research intent (”what’s best”), with REI showing up as the aggregator the models trust even at the research stage.

Crossover and media

Hyrox and functional fitness

1. Nike (69%)

2. Reebok (50%)

3. Adidas (34%)

4. PUMA (33%)

5. NOBULL (30%)

6. Under Armour (30%)

7. lululemon (25%)

8. On (24%)

Running media

1. Runner’s World (50%)

2. Trail Runner Magazine (29%)

3. UltraRunning Magazine (28%)

4. iRunFar (26%)

5. Strength Running (25%) - part of the Long Run Labs Podcast Network!

6. Women’s Running (20%)

7. Marathon Training Academy (19%)

8. Trail Runner Nation (19%)

The Hyrox board is a sponsorship cautionary tale. PUMA is the official title sponsor and sits fourth, while Nike, a non-sponsor, leads at 69 percent. The reason is timing: the models lag real-world reality by months because a recent deal has not yet saturated the training data or the live web. AI visibility is a trailing indicator of your marketing, not always a real-time one.

A real-world gut-check. Right as we were finishing this, journalist and Hurdle podcast host Emily Abbate posted from her first Hyrox event, and her on-the-ground read was the exact opposite of the model’s. She said PUMA totally owns this category, with athletes repping it everywhere, and even flagged LSKD as surging in popularity, a brand that does not appear in the model’s Hyrox top eight at all. That gap is the whole point. Emily is the leading indicator, reporting what is true today. The models are the lagging one, still describing the recent past until the roundups and coverage catch up. It is the cleanest illustration in this piece that owning a category in real life is not the same as owning it in AI. PUMA bought the sponsorship and the physical dominance. To win the AI answer too, the earned media has to say so first. When we re-run this study, PUMA should climb as the coverage saturates. Emily’s story today is the prediction. The next study is the proof.

Where the recommendations actually come from

When the models reach for a source at the category level, they reach for media, not brands. Runner’s World alone was cited in 40.5 percent of all 2,200 answers. Behind it sits a short, stable list (cited in this percentage of all prompts):

1. Runner’s World: 40.5%

2. iRunFar: 27.5%

3. OutdoorGearLab: 18.6%

4. RunRepeat: 14.7%

5. Treeline Review: 12.4%

6. REI: 12.1%

7. Outside: 12.1%

8. YouTube: 9.2%

The top ten domains account for a quarter of all 16,990 citations. Meanwhile every brand’s own website combined is under 5 percent of citations. Living on your own site is not a strategy. The models are reading a dozen magazines, and if you are not in those magazines, you are not in the answer.

They also go looking for those magazines by name. 7.5 percent of all AI fan-out searches named a specific publication:

• Runner’s World: 400 searches (4.5%)

• Wirecutter: 151 searches (1.7%)

• Outside: 107 searches (1.2%)

• OutdoorGearLab, RunRepeat, iRunFar: about 15 searches each (0.2%)

And there is a single dominant format. The “best of” list is the unit of recommendation (share of all citations):

• “Best of” and “top” roundups: 53%

• Reviews: 4%

• Guides: 3%

• News: 1.6%

• Reddit: 1.3%

• Wikipedia: 0.8%

The roundup is the currency, dwarfing everything else. Which means visibility in AI is won the way earned media has always been won: get into the roundups the models already trust, keep your story current because the models punish staleness, and publish your own educational content when you do publish. One underrated move: Wikipedia is ChatGPT’s single most-cited source across everything Centium has ever measured. You can contribute ethically (factually, no promotion), and at the research stage the models often read it more than your own site.

There is a quiet but strong winner in all of this: the earned-media shops (PR companies!). If the path into AI runs through the roundups, then the PR agencies that already live inside those publications just became a lot more valuable. For a decade, PR was the soft, hard-to-measure line item next to performance marketing’s clean ROAS. This data flips that. The pitch that lands your brand in a Runner’s World “best of” is now also the pitch that lands you in ten thousand AI answers. Earned media is becoming the most direct lever you have on AI visibility, and the people who are good at it are suddenly holding the keys.

What this means if you are a...

• Brand. Pick your category, get into the roundups, stay fresh, and know that it takes more content than you realize to move AI models meaningfully.

• Retailer. Intent flips the answer. “Best running shoes” surfaces media; “where can I buy” surfaces retailers, led by Road Runner Sports at 68 percent. Know which question you are trying to win.

• Race or event. This is real white space. The leader sits at 16 percent and 165 events surfaced. Nobody owns it yet.

• Tourism board. Perception is the product. Boulder won on a specific, repeated story, not on being objectively “best.” Pitch the activity-specific outlets (Runner’s World, Outside), not just the legacy travel press.

• Creator or media. Surface area is everything, because the model needs many places to find you before it trusts you. YouTube punches above its weight (the models index the transcript, not the video, so talk through your picks), and the format that travels is the ranked list. Quarterly “best of” pieces get picked up. Long narrative essays mostly do not.

The uncomfortable part

There is a real problem hiding inside all of this, and a recent guest, Christian Rawles, put it sharply: if the models strip-mine the content sites for their rankings, and runners get the answer without ever clicking through, what happens to the ad-supported publications producing the rankings in the first place?

It is not hypothetical. Michael described telling the editors at a major outdoor magazine that AI was leaning on them constantly, expecting them to be thrilled. Their response was closer to grief: the models are taking our work, and we see none of the traffic. Some sites are already blocking AI crawlers, which Michael thinks is a losing move, because invisible to AI increasingly means invisible, period.

Nobody has the answer yet. Maybe the roundups become a paid, vetted process, an application to be considered for “best gravel shoes of 2027,” the way other awards already work. Maybe the model makers eventually pay the publishers. What is clear is that today’s engine depends on a content economy whose business model AI is quietly eroding. That tension is the story to watch over the next two years.

The bottom line

AI did not hand running a level playing field. It built strong instincts about the giants, refreshes those instincts constantly from a small set of trusted publications, and rewards whoever shows up in the current, ranked, best-of content it goes hunting for.

But the door is wide open if you stop trying to win the whole sport and start trying to own your corner of it. Authority is earned category by category. The path in runs through earned media, not your homepage. And because the models chase what is new, this is not a trophy you win once. It is a habit you keep.

One thing this piece deliberately does not do is tell you how to actually get into those roundups. That is its own playbook, and it is what I am publishing next: Earned Media Is the New Performance Marketing, a tactical guide to winning AI visibility category by category. Subscribe and it will land in your inbox the day it goes live.

We are going to keep measuring this. The plan with Centium is to re-run the study every 6-12 months and track what moves, who climbs, and how the sourcing shifts. Consider this the baseline.

A quick note on For The Long Run

Yes, we checked how AI found my main/larger podcast, and it was a clean lesson in surface area. For The Long Run surfaced in five distinct prompts across ChatGPT, Claude, and Gemini, landing mid-pack at roughly 10 percent on the podcast question. The shows the models reach for most: Strength Running (63 percent), Marathon Training Academy (63 percent), Rich Roll (57 percent), Ali on the Run (50 percent), Trail Runner Nation (43 percent), Run to the Top (40 percent), and the morning shakeout (27 percent).

The interesting part was the mechanism. ChatGPT did not start with us. It researched the bigger shows first, found us listed in Apple’s “you might also like” panel next to Strength Running, Ali on the Run, and Trail Running Women, then ran its own searches on the show and confirmed us on our own site. We rode in on Apple’s “listeners also subscribed” graph. Two tells: ChatGPT’s write-up (”interviews exploring the why behind running”) is lifted almost verbatim from our Apple title, and on Feedspot’s “top running podcasts” list, which the models treat as close to ground truth, we sit at number 49. The fix is the same one the data prescribes for everyone: more surface area, a tighter and more consistent description everywhere the model might look, and a climb up the editorial roundups and aggregators.

---

FAQ

Does AI favor big running brands like Nike and HOKA?

On the broad question, yes. Ask a model for “the best running brand” and the incumbents return by reflex, because their instinct was built on a decade of the internet talking about the giants. Nine brands account for 22 percent of all mentions. But that grip is specific to the broad question. Go category by category and a different brand leads almost every one.

How do I get my brand recommended by ChatGPT and other AI models?

Get into the “best of” roundups the models already trust, own one category rather than the whole sport, and keep your coverage current, because the models reward freshness and punish staleness. Living on your own website is not enough. Brand-owned sites are under 5 percent of all citations.

Which sources do AI models trust most for running recommendations?

Media roundups, not brands. Runner’s World was cited in 40.5 percent of all answers, followed by iRunFar, OutdoorGearLab, and RunRepeat. The top ten domains account for a quarter of all 16,990 citations.

Why does winning one category not help in another?

The model cannot try on the shoes, taste the gels, or feel the watch, so for each category it goes to that category’s specialist sources. Authority does not transfer. Owning “best running shoe” and owning “best running headphones” are two unrelated wins.

Which AI models were tested, and how?

ChatGPT, Claude, Gemini, Perplexity, and Grok, across 22 categories at 20 unaided prompts each, for 2,200 total responses. Every prompt asked the category question and never named a brand, so the results measure organic visibility.

How often is the study updated?

The plan with Centium is to re-run it every 6-12 months. This is the June 2026 baseline.

Methodology

22 categories at 20 unaided prompts each, run across ChatGPT, Claude, Gemini, Perplexity, and Grok, for 2,200 total responses. 2,052 brands and entities surfaced, with 16,990 citations across 2,558 domains. Every prompt asks about the category and never names a brand, so the figures measure organic visibility. Results are aggregated across the five models (per-model counts are excluded because each runs under different output limits, making them non-comparable). Page-type classification is keyword-based, and the media-outlet share within fan-out counts a fixed list of named publications, so it is a floor, not a ceiling.

Source run: June 3, 2026. Study by Centium AI.

This piece is built from the full Centium AI study and my conversation with Michael Rueckert on Long Run Labs. Listen to the episode here (Apple) or here (Spotify).

If you want the practical, how-to companion, I wrote a three-part series on using AI as a marketer, founder, and operator, beginning here.

Want to know how your own brand shows up inside the models? Michael and the team at Centium AI measure exactly that, and there are free tools on their site to start.

A follow up to this piece can be found here:

---

Jon Levitt is the host of For The Long Run, founder of the Long Run Labs Network (35+ shows, ~1M monthly downloads), and co-founder of The Huddle. This newsletter covers the business of creator partnerships, sponsorship strategy, and what the data actually shows, in addition to a weekly article from that week’s Long Run Labs Podcast.