Profound vs AthenaHQ: Why Visibility Rose 63% While Conversions Dropped 90%
A B2B procurement SaaS watched its AI visibility rise from 13.1% to 21.4% while LLM-attributed conversions collapsed 90%. The pattern is real, and the cause is structural. Drawing on a KDD 2024 paper on Generative Engine Optimization, this review of Profound and AthenaHQ examines the eight engineering layers beneath the dashboard—robots.txt, llms.txt, schema, brand entity consistency—where commercial AEO tools deliver value, where they fall short, and why this gap keeps showing up.
A note on timeliness: This assessment of Profound and AthenaHQ draws on public materials available before June 2026 — funding announcements, customer case studies, third-party reviews. Both products ship fast, so check the official sources for the current state. This is not a buying recommendation.
An Illusion Shattered by Data
Between January and March 2026, a B2B client in procurement software watched a curve climb inside their AI visibility tracker — brand visibility rose from 13.1% to 21.4%, a 63% jump in three months. By the industry’s standard logic, where visibility equals citation rate, that curve should mean more prospects bumping into their name inside ChatGPT.
But over the same window, this client’s LLM-attributed conversions dropped 90%. Another enterprise SaaS client showed a subtler pattern: visibility edged up from 11.5% to 12.1%, AI citations were genuinely climbing, yet AI-referred traffic fell 64% — the brand kept getting mentioned, but users stopped clicking through. [1]
Both data points come from a public report published by the digital marketing agency Omnidiscient in May 2026. The report zeroes in on a pain point the AEO space rarely talks about in 2026: between the “visibility score” on the dashboard and the actual “AI citation conversion” sits a hidden disconnect. The score climbs; the business doesn’t.
The disconnect isn’t a tooling problem. It’s a measurement blind spot that runs across the whole industry. Dashboards measure the tip of the iceberg — prompt coverage, citation counts, share of voice. What actually decides whether an AI engine will reliably cite a brand, and whether users will bother clicking through, lies below the waterline: eight layers of foundational visibility engineering. None of them appear on anyone’s flagship features page, yet every advanced capability assumes they’re solid.
The sections below use those eight layers as a yardstick. They map the capability contours of Profound and AthenaHQ. But before unpacking them, the academic lineage matters — these aren’t a blogger’s checklist. They’re the engineering extension of a KDD 2024 paper cited more than 76 times.
GEO’s Academic Roots: The KDD 2024 Paper
In August 2024, Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, and Ameet Deshpande — from Princeton University, IIT Delhi, Georgia Tech, and the Allen Institute for AI — published GEO: Generative Engine Optimization at ACM SIGKDD, a top-tier data mining conference. It was the first large-scale, systematic academic study of how content gets cited by generative engines. [2]
The paper built the GEO-bench benchmark (spanning 25 domains and 10,000 queries) and tested nine content optimization strategies. The top finding: GEO methods can lift content visibility inside generative engines by up to 40%, with “adding citations” delivering +40%, “adding statistics” delivering +30%, and “adding quotation attribution” delivering +25%. A subtler result that most summaries skip: sites ranked lower (Google search position 5) gained 115% visibility through GEO, while top-ranked sites saw diminishing returns — GEO pays off most for those whose foundation is weakest. [2]
This paper gave the entire AEO/GEO space its academic legitimacy. Both Profound and AthenaHQ reference GEO concepts in their public materials, but neither has directly operationalized the paper’s “content visibility optimization strategies” as “foundational visibility audit dimensions.” The industry has loosely broken down that engineering base into eight layers — robots.txt, llms.txt, AI discovery, Schema JSON-LD, Meta tags, content quality, technical signals, brand entity — and these eight are the engineering implementation of the paper’s “credibility signals” idea. The rest of this piece walks through them side by side.
The Two Products in Profile
First, let’s put Profound and AthenaHQ in context. Both are top players in the 2026 AEO space, taking very different routes, and both qualify as industry benchmarks.
Profound bets on “data depth + content production.” According to public 2026 materials, the company has raised $58.5M to date, including a $35M Series B led by Sequoia Capital, with Kleiner Perkins and Khosla Ventures participating. The platform claims 500+ organizations and 2,000+ marketers as customers, processing a billion citations, 30 billion crawler visits per month, and 10 million prompts daily. Pricing starts at $99/month, and G2 sits at 4.6. Each account gets a dedicated engagement manager and AI strategist. [3][4]
Profound’s most compelling case study is Ramp. The finance automation company started at 3.2% AI visibility in the Accounts Payable category, ranked 19th in fintech. After running Profound’s Answer Engine Insights, Ramp shipped four targeted pages between December 2024 and February 2025, lifting citation share from 0.6% to 8.1%, generating 300+ citations across four pages (more than all prior content combined), and raising overall AI visibility 7x and overtaking 11 competitors. Another customer, Hone, posted an 800% visibility gain. [3]
AthenaHQ takes the “guided workflow + citation engine” route. It closed a $2.2M seed round in June 2025, led by Y Combinator, with FCVC, Red Bike Capital, and Amino Capital participating; SEO veterans Eli Schwartz and Ashley Stirrup came in as angels. Founder Andrew Yan is a former Google search engineer, and Alan Yao came from ServiceNow’s AI team. The platform has analyzed 3M+ real AI responses, mapped to 300K+ citing sites, with customers including SoFi, ZoomInfo, Wix, Coupons.com, and Checkr. Pricing uses a subscription-plus-credits model starting at $295/month, and G2 sits at 4.9. [5][6]
AthenaHQ’s flagship case is AutoRFP.ai. The Australian RFP automation software company posted 10x growth in ChatGPT referral traffic within eight weeks, with 30% of demo bookings originating from ChatGPT discovery and an 18-day payback period. Another customer, Verito, captured 36% share of voice despite a competitor with 25x their revenue. [5][6]
The market tailwinds are pushing both forward. McKinsey predicts 50% of consumers will use AI search by 2028, driving $750B in consumer decisions; Gartner forecasts a meaningful decline in traditional organic search traffic. AthenaHQ’s State of AI Search 2026 report has a sharper number: on average, only 17.2% of AI responses mention any specific brand, but top brands reach 56.7% — three times the market mean. BrightEdge fills in the rest of the picture: 83% of the content cited by AI Overviews comes from outside the traditional search Top 10. Translation: traditional SEO ranking and AI citation don’t move in lockstep — the paper’s “position 5 site +115%” finding gets market-side validation here. [6][7]
That’s the value anchor for both tools: helping you climb from 17.2% toward 56.7%. But the climb hinges on one assumption — the brand’s “foundational visibility base” has to be solid. The eight layers below are where the assumption meets reality.
The Eight Foundational Visibility Dimensions: Mapping Both Products
1. robots.txt: Letting AI Bots In
This layer checks whether robots.txt exists, and whether it correctly allows GPTBot, PerplexityBot, Google-Extended, ClaudeBot, Bytespider, and other mainstream AI crawlers. Full engineering practice also verifies the sitemap reference relationship and checks whether Disallow paths are accidentally blocking pages AI should see.
The real pain here is bigger than you’d think. Marketing.Chat’s 2026 GEO guide cites a number: 71% of publishers unknowingly block at least one crawler that feeds LLMs, losing AI search visibility without realizing it. Profound has a dedicated “AI crawlers” module that lists which AI bots are allowed and which are blocked, with fix suggestions — this is the entry point for its “infrastructure-grade crawler analysis” narrative. AthenaHQ’s Action Center also covers basic allow detection. Both treat this as part of “technical readiness,” not as a flagship selling point — the flagship features sit higher up the stack. [3][4]
2. llms.txt: The LLM-Specific Protocol
llms.txt is a 2024-era protocol, similar to robots.txt but aimed at large language models — it tells LLMs what content can be learned and what can’t. Full engineering practice includes: detecting whether /llms.txt exists, parsing whether the format follows spec, and verifying that declared resource paths are actually reachable.
An open-source GEO audit tool on Product Hunt, based on the Princeton KDD 2024 research, surfaced a number: fewer than 5% of websites have deployed llms.txt. As of public June 2026 materials, neither Profound nor AthenaHQ lists llms.txt as a primary detection dimension. This makes sense — the protocol is still in its early (v0.x) stages, and industry consensus hasn’t settled. But this dimension is already affecting real traffic. Both have chosen to focus on “mature mainstream AI bot allowlisting” rather than “emerging LLM protocol compliance” — a product roadmap decision, not a capability gap. [3][4]
3. AI Discovery: .well-known/ai.txt and /ai/summary.json
This layer gets the least attention of the eight. The AI discovery layer checks whether a site actively “introduces itself” to AI engines: /.well-known/ai.txt declares AI-accessible endpoints, /ai/summary.json provides a structured brand summary, /ai/faq.json serves up citation-ready Q&A pairs, and service.json describes service capabilities. Full engineering practice also validates the minimum length of name and description in summary.json, and the minimum length of each question and answer in faq.json — below the threshold, AI engines won’t use them even if they can read them.
This layer directly echoes the paper’s “adding citations +40%, adding statistics +30%” finding: AI engines prefer structured, verifiable, high-density information. Neither product has put this layer into its main detection flow in public materials. The logic is “AI will come find you” — they reverse-engineer brand visibility through prompt tracking and citation monitoring. That path works, and both do it at industry-leading levels. The AI discovery layer is the “active declaration” path, complementary to “passive measurement.” Both have chosen to invest engineering in the latter; the former is an industry-wide gap. [2][3]
4. Schema JSON-LD: Structured Data
This is a dimension both cover, and the one with the most mature industry consensus. Full engineering practice splits into four tiers: whether the structure exists, how many types are present, whether key types (like Organization/WebSite) are present, and whether key fields (like name/url/description/dateModified) are present.
The paper’s “adding quotation attribution +25%” strategy lands, at the engineering layer, on Schema field completeness — AI engines treat Schema as “machine-readable attribution.” Both Profound and AthenaHQ detect JSON-LD presence and type coverage; Profound also cross-analyzes schema against prompt data to judge which schema types are more likely to be cited by AI — an industry-leading data cross-referencing capability. Both are solid on “type presence”; on “field completeness,” public materials don’t show character-level validation details. Their core engineering strength is “intelligent analysis at the type layer”; field-level engineering validation is a different dimension. [3][4]
5. Meta Tags: title/description/OG/Twitter Card
This is a traditional strength for commercial tools. Profound gives optimization suggestions on title length, keyword density, and OG image dimensions; AthenaHQ’s Action Center can generate meta copy directly. Both lead the industry in “single-tag optimization.”
Full engineering practice also includes cross-tag consistency checks — whether og:title and twitter:title match, whether og:image and twitter:image point to the same resource, whether canonical and og:url align. This kind of character-level consistency check doesn’t appear in public materials for either. Commercial tools focus engineering on “generating better meta,” not “validating existing meta consistency” — the former is high-value incremental work, the latter is foundational structural work. Different positions, different jobs. [3][4]
6. Content Quality: Word Count, Heading Hierarchy, FAQ Density
Profound has a clear edge here — it runs a complete content production pipeline, generating new content and optimizing old content based on AEO data, one of the few platforms that bridge “measurement” and “production.” Those four targeted pages in the Ramp case were Profound pipeline outputs, directly lifting citation share from 0.6% to 8.1%. AthenaHQ’s Action Center surfaces content gap recommendations; in the AutoRFP.ai case, AthenaHQ helped identify 20% net-new content opportunities. [3][5]
This layer echoes the paper’s core strategies — “adding citations +40%, adding statistics +30%, adding quotations +25%” — the content quality audit layer essentially translates these academic strategies into detectable fields. But full engineering practice has a detail often missed in multilingual scenarios: word counting. Chinese counts characters, English counts words, and mixed-language content can’t be handled by a single regex. Neither product’s public materials show multilingual word-counting details; their corpus is English-first. Both are positioned for “content production in English contexts”; multilingual localization is another engineering layer. [2][3]
7. Technical Signals: lang Attribute, RSS, Freshness
This layer checks three things: whether the HTML tag’s lang attribute exists and is valid, whether an RSS/Atom feed exists, and whether schema or meta carries dateModified/datePublished. Freshness in particular has a meaningful impact on whether AI engines judge content “worth citing.”
Amsive’s 2026 research: 50% of AI-cited content is under 13 weeks old. Freshness isn’t optional — it’s a strong preference of AI engines. Profound’s technical SEO module covers lang and RSS; AthenaHQ’s public materials say less about this layer. Neither treats freshness as a standalone dimension; they frame it more under “content freshness” (whether it needs updating). Technical “is the update timestamp declared” and content “does it need updating” are two different dimensions — one is engineering config, the other is content ops. Both lean toward the latter because it’s closer to what customers notice. [3][4][7]
8. Brand Entity: Consistency, Knowledge Graph Anchors, About/Contact
This is the dimension where AthenaHQ’s ACE (Athena Citation Engine) focuses heavily, and it’s also a Profound strength. Both detect whether a brand is recognized as an “entity” by AI engines and whether it’s linked to a knowledge graph; AthenaHQ’s ACE also does citation probability prediction — one of the most technically substantive capabilities in the 2026 AEO industry.
Full engineering practice includes character-level validation of “brand name consistency” — H1, Title tag, and Schema name field all aligned. Plenty of sites write “Acme Tech” in H1, “Acme Tech Ltd.” in Title, and “Acme Technology” in Schema — three places, three variants, and AI engines can’t pin them as the same entity. As for finer checks — About link presence, Contact info, hreflang bidirectional backlinks — public materials show both focus on “whether it exists,” without character-level “whether it’s correct” detail. Their flagship capability is “predicting the probability of an entity being cited”; the foundational “entity consistency self-check” is the assumed premise — and whether that assumption holds depends on the site’s own engineering rigor. [3][5]
The Misalignment Between Advanced Capabilities and Foundational Engineering
With the eight layers mapped, the capability profile of both products comes into focus.
Profound and AthenaHQ concentrate their flagship capabilities “above the waterline” — prompt tracking, citation prediction, content production, Action Center. These are the 2026 AEO tool benchmarks, and the path from 17.2% toward 56.7% runs primarily through these advanced capabilities. Ramp’s 7x growth, AutoRFP.ai’s 10x traffic, Hone’s 800% lift — all are real cases of advanced capabilities paying off.
But the eight foundational visibility dimensions are a different layer — the engineering foundation “below the waterline.” Advanced capabilities assume the foundation is already solid: that robots.txt already lets AI bots in (while 71% of publishers unknowingly block crawlers), that Schema fields are already complete, that the brand name is already aligned across H1/Title/Schema, that freshness is already declared (while 50% of AI-cited content is under 13 weeks old). Once the assumption breaks, the “recommendations” from advanced capabilities break down in execution — that’s the root cause of the “visibility up 63%, conversions down 90%” data point. [1][7]
This isn’t a product defect in either tool; it’s the way the AEO industry splits responsibilities. The business model of commercial tools dictates that engineering investment sits at the advanced layer — that’s where customers directly feel value, and that’s where the two actually stand out. Foundational visibility engineering is a site’s own engineering responsibility; it shouldn’t, and won’t, be fully covered by any SaaS tool. The paper’s “position 5 site +115%” finding drives this home: the thinner the foundation, the bigger the payoff from fixing it. [2]
How to Check Whether Your “Below the Waterline” Is Leaking
Profound and AthenaHQ watch the waves on the surface for you, but the eight foundational visibility layers beneath need to be audited from an engineering perspective.
Before paying for an advanced AEO tool, do a free foundational check first. You can use GeoICU’s 18-dimension GEO audit radar to verify, in 30 seconds:
- Whether your
robots.txtis unknowingly blocking LLM crawlers (like 71% of publishers do) - Whether your
llms.txtis deployed correctly - Whether your Schema fields meet AI engines’ “attribution” standard
A weak foundation undermines everything built above it. Fix the base before buying the premium tier. 👉 Start your first free GEO foundational audit here
Closing
AthenaHQ’s ACE engine can predict the probability of a brand being cited; Profound’s content pipeline can generate content based on real prompts; Ramp used Profound to lift citation share from 0.6% to 8.1%; AutoRFP.ai used AthenaHQ to grow ChatGPT traffic 10x. These are top-tier capabilities in the 2026 AEO industry, and that’s why both deserve to be taken seriously. [3][5]
But before an AI engine cites a brand, it has to recognize it; before it recognizes it, it has to find it; before it finds it, it has to be able to crawl it. Pranjal Aggarwal et al. proved in the KDD 2024 paper that “credibility signals” can lift visibility by 40% — and the on-the-ground expression of that 40% is the eight foundational visibility dimensions: robots.txt access, llms.txt declaration, AI discovery endpoints, Schema fields, Meta tags, content quality, technical signals, brand entity consistency. None of these eight checkpoints will be covered for you by a commercial tool; each is the site’s own engineering responsibility. [2]
Advanced capabilities set the ceiling for brand visibility; foundational visibility sets the floor. When the floor leaks, no matter how high the ceiling, the water still rises around your ankles.
The score on the dashboard is an outcome, not a cause. The real causes live in the eight foundational visibility dimensions — in the engineering details that commercial tools assume you’ve already handled — and in that real story where “visibility rose from 13.1% to 21.4%, and conversions dropped 90%.” [1]
Sources:
[1] Omnidiscient: Cited, not clicked: The new math of AI traffic https://beomniscient.com/blog/why-your-chatgpt-traffic-just-fell-off-a-cliff/
[2] Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., Deshpande, A. (2024). GEO: Generative Engine Optimization. KDD 2024. arXiv:2311.09735. https://arxiv.org/abs/2311.09735
[3] SearchRoost: tryprofound.com Customer Success Stories: How Brands Win in AI Search https://searchroost.com/blog/tryprofound-customer-success-stories
[4] Profound official comparison: Profound vs. AthenaHQ: Which AI visibility platform is right for your brand? https://www.tryprofound.com/resources/articles/profound-vs-athenahq
[5] AthenaHQ official case study: 10x Growth in ChatGPT Traffic — AutoRFP.ai https://www.athenahq.ai/case-studies/10x-chatgpt-traffic-autorfp-success-story
[6] AthenaHQ official comparison: Profound vs AthenaHQ: The Ultimate AI Search Optimization Platform Comparison https://www.athenahq.ai/news/profound-vs-athenahq-comparison
[7] Marketing.Chat: GEO & AEO: the guide to showing up in ChatGPT, Claude, Gemini and Perplexity https://marketing.chat/en/guias/geo-generative-engine-optimization