Ahrefs

Ahrefs is how I learned to build a company. It's an SEO toolset — backlinks, keywords, content, rankings, site audits — but underneath it's a crawler and an index that we own end-to-end. Our own machines, our own data centers, over 100 petabytes of web data and more than $60M invested in our own infrastructure, no cloud bill in sight. The link graph that comes out the other end — 35 trillion backlinks from 500 million referring domains, which a 2026 head-to-head comparison calls “the industry's largest index of referring domains” (Semrush trails at 390M) — is what marketers actually pay us for. The keyword side is the same shape: 28.7 billion keywords across 243 countries in the database — a corpus 287× the size of Ubersuggest's and measurably ahead of Semrush's 26.7B. The product gets better and cheaper every year because we control the stack.

“Fresh index, good coverage, not get blocked. Choose two.” — the line I use to explain why building a real web index is hard, published on Ahrefs Academy · Crawl budget

Today, roughly 50,000 paying customers rely on Ahrefs every month — from solo freelancers and small agencies to teams at Facebook, eBay, LinkedIn, Adobe, Netflix, and Uber — and ahrefs.com pulls in around 20M+ monthly visits. The same 2026 head-to-head that calls our referring-domain index the industry's largest puts the commercial picture in the same shape: ~54,215 companies actively use Ahrefs vs. 24,854 for Semrush (a roughly 2:1 ratio of enterprise adoption), and per the same article's reading of 6sense's 2026 SEO/SEM market analysis Ahrefs holds 14.83% category market share against Semrush's 6.68% — about 2.2× the share of our closest competitor. And we've never had a sales team. What we have instead is a content engine: Ahrefs' Digest reaches 284,000 weekly newsletter subscribers — the number we publish ourselves in our own roundup of the SEO newsletter landscape — sitting on top of a blog that pulls in roughly 600K monthly visits and earns its trial signups directly from search, not from outbound. The compounding shows up in the topline: $37M (2018) → $65M (2020) → $86M (2021) → $100M (2023) → $149M (2024) — roughly 4× revenue in six years, no funding rounds, no dilution, no sales team to feed it. Customer + traffic figures per inblog.ai's Ahrefs growth case study (2025); revenue trajectory per Latka's Ahrefs financial profile; newsletter subscriber count first-party from ahrefs.com/blog/seo-newsletters (“Ahrefs’ Digest … Number of subscribers: 284,000”); blog monthly-visits figure per yespress.io's Ahrefs blog profile.

How that lands with the people who do this for a living: in Aira's State of Link Building Report 2022, 64% of SEO professionals said they trust Ahrefs' link data over any competitor — the next closest, Google Search Console, got 14%. 67% of digital marketers reach for our Domain Rating when sizing up a backlink (vs. 42% for the next metric). Source: ahrefs.com/blog/ahrefs-seo-metrics, citing Aira's State of Link Building Report 2022; DR figure per PressWhizz, same report.

“It took years of effort. We didn't have the money that Google or Microsoft have, so we optimized everything to work with minimal costs. Google's index is 26 years old, Bing's is 14 years old, and ours is 13. We can technologically and morally compete with giant companies on volume, infrastructure, and developer expertise.” — me, to Newssky

Agent A. Our biggest shift in years — Agent A, the AI marketing agent built on top of Ahrefs' full dataset (170 trillion+ indexed pages, by our own count), launched in 2026. It's not a wrapper that queries a filtered API — it's an agent with unrestricted access to the underlying index, plus pre-built marketing skills trained by our own team. Connect your GA4, GSC, Google Ads, Meta Ads, WordPress, Shopify and it runs analyses, builds reports, and acts on its own. The thesis behind it, in one line:

“It's time to delegate that to AI — so you can operate on a higher level of abstraction and focus on what actually matters.”

That's the founder's framing; here's an independent one. In ALM Corp's 2026 roundup of 12 popular SEO AI agents — Agent A sitting alongside Frase, NightOwl, AirOps, n8n, and the rest — the reviewers land on the same thesis I tried to ship: “A common failure in AI SEO work is that the model writes fluently but reasons weakly because it is not operating from a strong SEO dataset. Agent A addresses that problem by using the depth of Ahrefs data as the foundation for its outputs.” That's the whole bet, told back to us by someone who doesn't work here. Source: ALM Corp, “12 Popular SEO AI Agents in 2026.”

Why we built this, in the same plain language we put in the press release: “As AI assistants become a mainstream discovery channel, teams are increasingly trying to understand whether their brand shows up in AI-generated responses at all.” So in January 2026 we shipped custom AI prompt tracking in Brand Radar — so teams can see how their brand actually shows up in ChatGPT, Gemini, Perplexity, and Google AI answers. Underneath it sits one of the largest AI-visibility datasets anywhere: 239M+ prompts indexed every month (162M AI Overviews + 29M AI Mode + 100M across ChatGPT, Copilot, Gemini, and Perplexity). The combined platform now reaches 44% of the Fortune 500. Then in December 2025 we extended the same infrastructure beyond AI answer surfaces: Brand Radar now indexes YouTube, TikTok, and Reddit too — scanning video titles, descriptions, and transcripts for brand mentions, plus tracking how Reddit threads show up inside Google search results. Tim Soulo, our CMO, on why: “ChatGPT and other AI tools are becoming an important discovery layer, but they're not where demand starts. Demand is created on platforms that capture massive attention — YouTube, TikTok, and Reddit — where people talk, share, and influence each other.” Which is the practical version of a finding our own research team measured and is reported a couple of paragraphs down: YouTube is the single strongest correlate of brand visibility in AI answers, and the #2 most-cited domain in AI Mode after Wikipedia. We studied where the citations come from; we shipped the tool to track them.

How we shaped it differently from the rest of the AI-visibility market, in Tim Soulo's own writeup comparing Brand Radar to Profound (one of the first-mover GEO dashboards, Fortune-500-priced): “I wanted Brand Radar to be an AI SEO research engine, not just a monitoring dashboard. And I wanted the good stuff accessible without an enterprise contract.” Profound gates full-engine coverage (Claude, Grok, Rufus) and ChatGPT Shopping behind Enterprise; their Starter plan is one engine and 50 prompts. Brand Radar puts all six engines (AI Overviews, AI Mode, ChatGPT, Gemini, Perplexity, Copilot) and the 239M-prompt dataset on the same shelf as our regular SEO product — same logic as $99 Agent A, free Yep API signup, free Firehose beta: the good stuff doesn't sit behind a sales call. Source: Tim Soulo, “Profound AI Alternatives: Why I Think Ahrefs Brand Radar Is the Smarter Choice” (Medium, 2026); engine-list per Brand Radar's own product page and the Jan 2026 BusinessWire release.

What that looks like in production: Octopus Energy — the UK energy retailer with ~7M+ customers — replaced a resource-heavy manual process for tracking AI visibility across global markets with Brand Radar, and used the output to brief their C-suite and reshape their content strategy for AI search. Laura Lyons, their SEO lead, on the platform: “If you want a really intuitive platform for tracking and understanding your AI results, one that's easy to use, easy to explain, and useful for both SEO and non-SEO teams, this is it.” Source: Ahrefs' published case study, “How Octopus Energy uses Ahrefs Brand Radar to monitor AI visibility across global markets”.

And at the other end of the buyer spectrum, on the agency side of that same Fortune-500 line: David Stein, SVP of Performance Marketing at VML — one of the world's largest marketing agencies, ~30,000 people across 60+ markets — on our public reviews page: “Ahrefs has been the number one tool in our SEO tool belt for many years. From the basics like rank tracking to the pro-level backlink insights, Ahrefs allows us to be more efficient and more effective SEOs.” Two named voices, one utility serving millions of households and one agency serving Fortune-500 brands — the 44% number isn't a slide; it's two distinct buyer personas saying the same thing in their own words.

That dataset already produces findings nobody else can. In a study of 75,000 brands, our team found that YouTube mentions are the single strongest correlate of brand visibility in AI answers (~0.737) across ChatGPT, AI Mode, and AI Overviews — outperforming Domain Rating, branded search volume, and raw content volume. A second study, this one analyzing 5.5 million Google AI Mode queries, found the same shape from the other direction: YouTube is the #2 most-cited domain in AI Mode answers (behind only Wikipedia, ahead of Reddit and Google's own properties) — two independent datasets converging on the same uncomfortable conclusion for the SEO playbook of the last decade. In February 2026 we measured the cost of the new SERP directly: comparing position-one CTR for informational queries in December 2023 (pre-AI-Overviews, 0.076) against December 2025 (0.039), we found AI Overviews now strip 58% of clicks off the #1 organic result — a measurement of what these answer surfaces actually take from the open web. And in May 2026, Louise Linehan and Xibeijia Guan tracked 1,885 pages that added JSON-LD schema against 4,000 matched controls and found AI citations barely moved (AIO −4.6%, AI Mode +2.4%, ChatGPT +2.2% — the latter two statistically indistinguishable from zero), pushing back on a year of folk wisdom that schema is the AI-citation cheat code. The deeper decoupling is the one we measured in March 2026: across 863,000 keyword SERPs and 4 million AI Overview URLs, only 37.9% of AI-cited pages also ranked in Google's top 10 for the same query — down from ~76% in July 2025, eight months earlier. The remaining citations split almost evenly between positions 11–100 (31.2%) and pages beyond position 100 (31.0%). Two out of three AI Overview citations now come from pages that don't rank in the top 10 of the same query they're cited for. Ranking and being cited used to be roughly the same goal; in eight months they've come apart. The other half of the same finding, measured by people outside Ahrefs: AirOps's 2026 State of AI Search report found 85% of brand mentions in AI-generated answers come from third-party pages — not the brand's own site — with McKinsey's parallel research putting brand-owned domains at just 5–10% of the sources AI assistants pull from. So the rank-to-citation decoupling has a destination: AI is citing what other people write about you, not what you write about yourself. Which is exactly the measurement Brand Radar was built to give back to brands, and exactly why the AI-visibility playbook now leans on earned mentions, reviews, and forum posts the way the link-building playbook of 2015 leaned on guest posts. We did the same exercise on the AI-content question itself: a study of 600,000 pages (100,000 random keywords × top-20 results) measured the correlation between how much AI text a page contained and where it ranked on Google — the answer was 0.011, statistically indistinguishable from zero, with 81.9% of top-ranking pages using a human/AI blend and only 13.5% purely human-written. Whatever Google is rewarding, it isn't authorship. And on the demand side — the question publishers ought to be asking once they accept the click loss is real — Patrick Stox measured AI-search traffic against organic search on ahrefs.com over a 30-day window and found something the click-loss panic obscures: AI-search visitors are just 0.5% of total traffic but drive 12.1% of signups — a 23× higher conversion rate than traditional organic search. Less traffic, much better traffic. Eleven months later, Mateusz Makosiewicz replicated the same finding at fleet scale: across the 74,752 sites we track, AI chatbots drove 3.5M visits in March 2026 — exactly 0.28% of total web traffic, with ChatGPT at 2.7M and Perplexity and Gemini at ~230K apiece — while Google still sent 345.2M. But Google's share is moving: 35.11% (June 2025) → 30.53% (March 2026), and the conversion-rate gap is the same ~23× the single-domain Stox study found. Single-site result, repeated at the scale of 74K sites a year later — the click-loss panic is real, but so is the much-better-traffic finding underneath it. Adobe's enterprise data arrives at the same shape from the other side (5% conversion lift, 12% engagement lift on AI-referred visits). The web is being rebuilt around fewer, higher-intent arrivals — which is also, not coincidentally, the deal an honest open web ought to want.

One concrete face on the abstract number: on January 8, 2026, Tailwind CSS laid off 75% of its engineering team — from four engineers to one. The framework is used by NASA, Netflix, and roughly 617,000 other sites; documentation traffic is down ~40% since early 2023 because AI-generated search answers now resolve the queries that used to land on Tailwind's docs, and the docs were the discovery funnel for the paid products. Creator Adam Wathan, in his own words on the GitHub thread where the news broke: “Tailwind is growing faster than it ever has and is bigger than it ever has been, and our revenue is down close to 80%.” That sentence is what 58% click strip and 0.28% chatbot traffic and 37.9% rank-citation overlap actually look like in one specific business that ships every day. It's also why the rest of this card stops being abstract from here: an index that turns into a platform, a streaming firehose, an agent that fixes pages, a free bot-traffic dashboard for publishers — the work for the next decade is rebuilding the deal between the open web and the answer surfaces eating it. The Tailwind story is the deadline. Sources: Search Engine Roundtable (Barry Schwartz, Jan 8 2026, citing Adam Wathan's GitHub post) and Slashdot's Jan 8 2026 writeup; verbatim quote from Wathan's own post in tailwindlabs/tailwindcss.com#2388.

We've also opened the index in the other direction. In September 2025 we launched Ahrefs Connect — an OAuth-based integration program on API v3 that lets third-party tools (Screaming Frog was one of the first verified apps) pull backlinks, organic search, Site Audit, and Rank Tracker data on behalf of a user, paid for in API units from their own subscription. Our remote MCP server does the same job for AI assistants: a single hosted endpoint at https://api.ahrefs.com/mcp/mcp that any MCP-compatible client can speak — we're listed in Claude's official Connectors directory, and the Claude Code install is one line (claude mcp add ahrefs https://api.ahrefs.com/mcp/mcp -t http). Ahrefs data straight inside Claude, ChatGPT, Cursor, and the rest — SEO research as a conversation, not a dashboard. And in early 2026 we shipped the most agent-native opening yet: Firehose, a real-time web-data streaming API that turns AhrefsBot's continuous crawl into a live SSE feed — define a Lucene rule (title:tesla AND page_category:"/News"), and the moment a matching page is crawled, your agent gets pushed an event with no polling. An independent developer at Morgen used the beta to build 5 marketing tools as plain Node scripts — “no databases, no servers to maintain” — which is roughly the point: the crawl infrastructure we own is becoming a primitive other people compose on top of. The index is becoming a platform other people get to build on. SEO is changing shape; the index has to keep up. Firehose first-party docs and pricing: firehose.com (“free during beta, no credit card”); independently listed in Searchbloom's 2026 AI Search Optimization tools guide as “Ahrefs Firehose … free during beta as of launch (March 2026)”; builder writeup from the Morgen team on Substack.

And in 2026 we closed the loop on the other side: Patches by Ahrefs takes the issues Site Audit already finds and lets you fix them directly from inside Ahrefs — “no developers needed”, in the product's own words. Titles, meta descriptions, canonicals, redirects, noindex/nofollow tags, image alt text, internal linking: AI drafts the fix, you approve, Patches publishes — either via a one-line JavaScript snippet on the page, or as a Cloudflare Worker that rewrites server-side before the page is delivered. Same idea as Firehose, opposite direction: if Firehose makes the index agent-readable in real time, Patches makes your own site agent-writable without touching the CMS. Agent A reasons over the data; Firehose streams the data; Patches acts on the page. The full agent-native loop, all on one bill. Source: Ahrefs' own Patches product page (the eight patch-supported surfaces, “no developers needed,” JavaScript-snippet vs Cloudflare-Worker deployment modes, and Patches' inclusion in the Project Boost Max add-on are all stated there); add-on context and Batch AI / IndexNow auto-submit per the Ahrefs Project Boost help doc.

In September 2025 we did something we'd never done before: acquired Detailed.com and the Detailed SEO Extension (450k+ weekly Chrome users), and brought founder Glen Allsopp on full-time. Bootstrapped doesn't mean we never buy — it means when we do, the money is ours and the call is ours. And here's the ship-velocity that buys: in the first four months after Glen joined, he counted 85 product updates the team shipped — his words, on his own blog, not ours: “I’ve been an Ahrefs customer for more than ten years, I never realized quite how often new features are added to the platform.” Roughly 21 shipped updates a month, sustained — the version of “decade thinking” that doesn't look slow up close. And the cadence hasn't slowed: our public changelog is a dated, running list anyone can scroll — the most recent batch alone (Sept 2025–Jan 2026) names individual ship dates like Dec 17: custom prompt tracking in Brand Radar, Dec 17: AI-classified page types in SERP overview, Dec 11: historical GA backfill in Web Analytics, Sept 18: MCP available on Lite+. The receipt isn't Glen's count or ours; it's a public log with dates. Acquisition per Search Engine Journal (Sept 2025); 85-updates-in-4-months count and the “more than ten years… never realized” quote per Glen Allsopp's own “Custom Prompt Tracking, and 9 Other New Ahrefs Features I Love” on Detailed.com; running ship log + dated Sept–Jan 2026 entries per the first-party Ahrefs changelog at ahrefs.com/blog/new-features.

We also show up in person. Our home-turf conference Ahrefs Evolve is built around one deliberately honest question — in our own words from the 2026 GEO-conference roundup we published, it's “600+ people from 35+ countries who are all wrestling with the same existential question: how do we stay discoverable when AI is eating our clicks?” The 2026 theme is named the same way: “staying discoverable in search, AI, and beyond.” Singapore wrapped on May 14, 2026 (Tim Soulo headlining, with Sasha Gusain of Canva, Dan Petrovic of DEJAN, and James Norquay of Prosperity Media); next stop is San Diego, October 12–13, 2026 — a single-track, no-pay-to-play lineup that includes Mark Schaefer, Cyrus Shepard, Sam Oh, Peep Laja, Kevin Indig, Patrick Stox, Emily Kramer, Ryan Law, and folks from IBM, Microsoft, HubSpot, and Brainlabs. If you want the unedited version of any of this, that's where to find it. Thesis quote + format details (600+ people, 35+ countries, single-track, no pay-to-play) per Ahrefs' own 14 GEO Conferences to Attend in 2026; Singapore lineup per Ahrefs' content-marketing-conferences roundup; San Diego speaker list per The Masterminders' 2026 SEO-conferences comparison.

ahrefs.com → · ahrefsevolve.com →

Yep

Why “Yep”? — “It doesn't have any specific meaning but is a nice, short, and easy-to-remember name. Choosing a name for the search engine was quite a struggle. We checked hundreds of ideas in the last 2 years. Initially, the team settled with Fairsearch dot com, with plans to transition to Fair.com. But we couldn't get Fair.com. Yep itself came up the first time when we were watching the Avatar cartoons in which Aang & his friends used the phrase, ‘yip-yip’ to make their sky bison take off into the air.” — me, in Search Engine Journal.

Yep started in 2019 as a swing at a creator-funding search engine — share 90% of ad revenue with the people whose content actually answers the query. Then AI Overviews and ChatGPT rewrote how people find things, and the revenue model that worked for YouTube stopped working for the open SERP. So we changed shape.

1. 2019
2. Announced Yep as a new search engine pursuing a 90/10 creator revenue split — the original swing at a fairer search ecosystem.
3. ~2020
4. LLMs, AI Overviews, and ChatGPT change how information is found. The ad-share model designed for the old open SERP no longer fits the new shape of search.
5. 2025
6. Refocused Yep around a Search API — same 100B-page index, same independence, now purpose-built to ground AI products in fresh web data.

Timeline per platform.yep.com (Yep's own published history).

The fair version of what happened to 90/10: it was never abandoned by us, it was outrun by the shape of search. The model only works at scale — ad revenue per query has to be meaningful, and there have to be enough queries. Yep never reached that scale as a consumer destination before generative AI changed where queries went in the first place. An honest 2026 outside review put it bluntly: “90% of very little is still very little.” The principle didn't go away — it migrated. A Search API for AI is how an independent, non-Google/non-Bing index of the open web gets paid in 2026, and every dollar that flows through Yep instead of through a Google-proxy is a dollar that funds a crawler that pays publishers in attention rather than in lock-in.

“Creators who make search results possible deserve to receive payments for their work. We saw how YouTube's profit-sharing model made the whole video-making industry thrive. Splitting advertising profits 90/10 with content authors, we want to give a push towards treating talent fairly in the search industry.”

The other founding pillar was privacy. From day one Yep was designed to collect the minimum data needed to return a result, and to never tie it to a person — “we do save certain data on searches, but never in a personally identifiable way.” No location, no age, no profile of you sitting behind the query. A small thing to promise; a structurally hard thing to mean, because the standard search business model is built on the opposite. Dmytro quoted in Yext's writeup of the Yep launch, naming privacy and profit-sharing as the two problems Google “will never want to fix.”

Today Yep is a Search API for AI: 100 billion pages in the index and news stories (per platform.yep.com's own hero — news is called out as a distinct corpus, not bundled into “web pages,” because fresh news is the hardest content for a stale-index API to serve well and the most valuable for an AI assistant to ground on), 8 billion pages crawled daily by AhrefsBot, one API call for grounding LLMs in fresh web data. Same index, same independence, new job. Pricing is the kind of thing a bootstrapped shop can do: $4 per 1,000 requests (Balanced) or $8 (Advanced), pay-as-you-go, no commitments, free signup — not “contact sales.” The demand side is showing up: at HubSpot's April 2026 Spring Spotlight, the company reported its customers' organic traffic fell 27% year-over-year while AI referral traffic tripled — the exact substitution Yep was built to ride. A small follow-through receipt while we're here: in the 2022 TechRadar Pro launch interview I said “our algorithm works well for short queries, but we still have to improve longer queries. We also want to have images, news and video.” Four years later, yep.com's nav has All / Images / News tabs shipped, longer-query quality is the headline item on our public roadmap (“improving search result quality for text search, image search and News”), and video is the one still outstanding. Two out of three, on the record, in public.

Why this matters now: on August 11, 2025, Microsoft retired the Bing Search APIs — the second-largest independent web index, gone from the developer market overnight. Teams building AI features that need fresh, broad web data suddenly had one fewer place to turn. There are very few independent, full-web indexes left you can hit with an API call: by published numbers, Brave Search calls itself “the third-largest global independent search engine, with an index of over 30 billion webpages” — Yep's 100B pages is roughly 3.3× that, on the same open-web, no-Google/Bing-proxy footing. We're one of those independents, and we plan to stay one. And the industry press has started to read the field the same way: in Search Engine Journal's February 2026 ranking of 25 alternative search engines, Yep places #5 overall — behind only ChatGPT Search, Google AI Mode, Perplexity, and You.com, and ahead of Brave, DuckDuckGo, Ecosia, and every other named alternative on the list. The shape of the top is “AI-grounded answer engines plus the independent indexes that can ground them,” which is exactly the category we're now built to serve. Sources: Microsoft Lifecycle — Bing Search APIs retirement announcement; Brave's own AI Grounding launch post (self-reported 30B-page index, third-largest independent); Search Engine Journal's “25 Alternative Search Engines You Can Use Instead Of Google” by Chuck Price, Feb 14 2026 (Yep listed at rank #5).

Latest: Yep now has its own crawler. YepBot ships as a dedicated user-agent (Mozilla/5.0 (compatible; YepBot/1.0; +http://yep.com/yepbot/)) — still backed by AhrefsBot as the primary source today, but built to take over more of the crawl as Yep's role as a search API for AI grows. Publishers get a single, documented bot to allow or disallow; we get room to evolve Yep's index on its own clock.

And a credential that doesn't get said enough: Yep is an official IndexNow participant — the open protocol publishers use to push real-time “this URL just changed” signals to search engines. The other names on that list are Microsoft Bing, Yandex, Seznam, Naver. We're one of a handful of independent engines that publishers can notify directly when their content updates, instead of waiting for a crawler to find it. “Fresh index” isn't just a crawl frequency; it's a two-way protocol the rest of the open web has agreed on — and we showed up to sign it. Source: ahrefs.com/robot (first-party) and the IndexNow participants list.

yep.com → · platform.yep.com → · yep.com/yepbot →