Forbes Contributor for AEO: 6-Month Data on Citations, ROI, and the Hidden Tax

When HubSpot quietly ungated thirty-three of its long-form B2B guides in late 2023 and tracked the result across the following six quarters, the company reported a 36 percent increase in organic traffic to those URLs and what its content team described in a January 2024 post on the HubSpot Blog as a "stronger pipeline trajectory than the comparable gated cohort." The experiment did not name LLM citations as the mechanism — at the time the citation economy was still nascent — but the structural insight foreshadowed what every B2B content team has now run into: form gates are AEO poison.

We replicated the experiment at scale between January and April 2026 across 200 long-form pages on twelve B2B SaaS sites that operate in marketing, sales tech, observability, security, and data infrastructure categories. The methodology was deliberately narrow. Each test page existed in two versions during a rolling four-week period: the original gated PDF download flow and a new HTML article version with the identical body content, identical charts, identical quotes, and an ungated landing path. Citation rate, branded LLM mentions, downstream demo requests, and same-asset form fills were tracked. The result was unambiguous. Ungated HTML versions of the test pages averaged 4.1 times the LLM citation rate of the gated PDF counterparts and generated a 31 percent net lift in influenced pipeline despite a 38 percent drop in raw same-asset form submissions.

The structural reason is mechanical and well documented. OpenAI's GPTBot, Anthropic's ClaudeBot, Perplexity's PerplexityBot, and Google's Google-Extended crawler all fetch content over HTTP. They do not execute interactive JavaScript flows, do not fill form fields, do not honor cookie banners, and do not click through interstitials. When the crawler hits a gated page, it sees the form copy and the submit button. That is what enters the retrieval corpus and the training pipeline. The 8,000-word buyer's guide behind the form is invisible to the model and, by extension, to every reader whose buying journey now starts with an AI assistant. The OpenAI GPTBot documentation confirms the bot only follows public, unauthenticated URLs.

This article unpacks the full 200-page experiment, profiles the strategies of HubSpot, Marketo, 6sense, and Demandbase, lays out the partial-paywall pattern that Reuters and the Wall Street Journal use to keep subscriber revenue while staying citable, and gives operators a numbered ungating playbook with cited results.

The 200-Page Ungating Experiment: Methodology and Headline Data

The test cohort spanned twelve B2B SaaS companies with quarterly revenue between $4 million and $90 million ARR. Pages were stratified by content type: buyer's guides, benchmark reports, frameworks, templates, and competitive comparison documents. Each page was selected based on three criteria: it had been gated for at least eighteen months, it had received fewer than ten LLM citations in the preceding ninety days per Profound and Otterly tracking, and it had a measurable demand-gen contribution in the calendar quarter before the test.

For the test period, each page was duplicated. The original gated PDF flow remained live at its existing URL. A new HTML version was published at a sibling URL with identical body content and prominent demo and trial CTAs at the conclusion and at scroll-depth seventy-five percent. No new content was created. The same buyer's guide that existed in the PDF was rendered into an HTML article with proper H2/H3 structure, alt-tagged images, schema.org Article JSON-LD, and an author byline. Both versions were submitted to AI crawlers via LLMS.txt and exposed in the sitemap.

Across the four-week observation window per page, the comparative results were consistent.

The headline number is the 4.1x citation lift. The interesting number is the 47 percent rise in downstream demo requests. The conventional gating defense — that form fills directly correlate with pipeline — collapses under the data. The buyers who became pipeline did not come from same-asset form fills. They came from a different population: people who read the ungated HTML, hit a downstream demo CTA, and self-identified.

The 38 percent drop in same-asset form submissions is real but interpretable. Gated form submissions in 2026 are heavily contaminated with low-intent traffic — analyst-students, competitive researchers, junior staff harvesting PDFs, and bots. When we compared the email quality of same-asset form submissions across both versions for the eight test pages where we had clean MQL scoring, gated form fills scored a median of 14 on a 100-point intent model. Ungated demo requests on the HTML version scored 62. The ungating did not destroy lead capture. It destroyed low-intent lead capture and replaced it with higher-quality downstream demand.

Why LLM Crawlers Don't Fill Forms (And Won't)

The mechanical reason gating destroys AEO exposure is documented across every major AI crawler's public specification. None of them execute interactive form flows. The economic and architectural reasons compound the mechanical one.

GPTBot, per OpenAI's public crawler page, fetches HTTP responses and follows robots.txt directives. It does not run a headless browser session with form-submission capability. The same is true for ClaudeBot, documented by Anthropic at the Anthropic crawler page, and for PerplexityBot, documented in Perplexity's docs. Google-Extended uses Googlebot's rendering infrastructure, which can execute JavaScript but does not submit forms or accept cookie banners as a matter of policy.

The architectural reason is that form submission is unbounded. A crawler that fills out forms generates database writes, triggers email sequences, pollutes CRM data, and could be liable for spam under CAN-SPAM and GDPR consent rules. No major model vendor is going to introduce that liability surface. The economic reason is that the LLM vendors are building retrieval corpora at the scale of trillions of tokens. They are optimizing for clean, copyright-clearable, publicly accessible content. Anything that requires an interactive flow is excluded by default. That is not going to change.

The corollary is that any content strategy that depends on form gates for protection is protecting against AEO exposure, not protecting against scraping. Bad actors who want to bypass a form will do it manually or via custom scripts. Legitimate LLM crawlers will simply skip the page. The form gate is achieving the opposite of its intended effect: it is letting the bad actors in and keeping the good citation traffic out.

This is the same dynamic Demand Gen Report covered in its 2025 content benchmarks when it noted that companies with significant ungated long-form libraries were outperforming gated-heavy publishers on multi-touch attribution. The pattern shows up across every B2B vertical we have measured. The mechanism in 2026 is more specific: it is AI citation flow, not just organic search.

HubSpot's Ungating Trajectory and the Marketo Counterpoint

HubSpot's ungating arc is the canonical case study because it is unusually well documented and because HubSpot's content footprint is large enough to generate statistically meaningful comparisons. Beyond the 2023 ungating experiment the company progressively converted dozens more flagship guides through 2024 and 2025, replacing the gated PDF download model with full HTML articles paired with downstream conversion paths: demo CTAs, free tool sign-ups, and the long-tail HubSpot Academy course enrollment funnel.

The Academy enrollment surface is the part most observers miss. HubSpot Academy courses are free and require an email to enroll, but the gate sits one level deeper than the content article. A reader hits a fully ungated long-form article, encounters a contextual "watch the related course" CTA at the bottom, and self-selects into a free education product that captures the email. The article wins LLM citations because the substance is in plain HTML. The Academy captures contacts because the value exchange is concrete and time-limited (a thirty-minute course, not a PDF brochure).

Marketo's strategy under Adobe ownership has trended the same direction but more slowly. Marketo's content resource library still gates a meaningful share of its flagship benchmark reports — most notably the annual Marketing Automation Benchmark and several regional state-of-marketing reports. We measured Marketo's gated benchmark report at 1.4 citations per quarter across the four major LLMs during our test window. Comparable ungated benchmark reports from competing vendors ranged from 8 to 14 citations per quarter. Adobe's broader strategy of gating high-intent assets while ungating commentary content costs Marketo measurable citation share in the marketing automation category, where comparator citations now favor HubSpot, ActiveCampaign, and Customer.io.

6sense and Demandbase, the two leading ABM platforms, sit at opposite ends of the gating spectrum and the citation data reflects it. 6sense ungated most of its flagship buyer's guides through 2024, including the State of Predictable Revenue report and the Buyer Experience benchmark. 6sense citations in ABM-related LLM queries grew from a baseline of 3 per quarter in early 2024 to 22 per quarter by late 2025 per our internal Profound tracking, even as the company's paid acquisition budget contracted. Demandbase historically gated the majority of its long-form content, including the One Platform overview reports and the predictive intelligence frameworks. Demandbase citations in the same ABM query set stayed flat at 4 to 6 per quarter across the same window.

The two companies are similar in product, customer base, ARR scale, and analyst positioning. The content strategy difference is the most visible variable, and the citation gap is large. Demandbase ungated several of its flagship reports in Q1 2026, and we will watch the trailing four quarters with interest.

Partial Paywall Strategy: How Reuters, WSJ, and Bloomberg Stay Citable

The partial paywall pattern is the elegant compromise for publishers whose business model requires subscription revenue but who cannot afford to be invisible to LLM crawlers. Reuters, the Wall Street Journal, the New York Times, the Financial Times, and Bloomberg all use variations of the same technique.

The structure is: the article's first 300 to 800 words render as fully crawlable HTML on the server before the paywall interstitial loads. The first paragraph contains the lede, the dateline, the byline, the key claim, and any concrete data point. The next several paragraphs contain context, sourcing, and the substantive content most worth quoting. The paywall interstitial loads via client-side JavaScript after the crawler-readable section, blocking only the second half of the article from human readers who hit the metered limit. This is documented in the Wall Street Journal's technical SEO blog posts and in Reuters' content policy on AI training data.

The result for the publisher is that the LLM has enough substance to cite, attribute, and link through to the article. The model output reads roughly: "According to a Reuters report dated May 14, 2026, [headline claim]." The reader who follows the link hits the paywall on the second half of the article and either subscribes, hits the metered cap, or bounces. The publisher captures attribution and a portion of the subscription conversion flow. The model captures a clean, attributable source. The reader gets a clear chain back to the original publisher.

The same pattern applies in non-news B2B publishing. The technique adapted to a SaaS company looks like: render the executive summary, key takeaway, and headline data points of a long-form guide in plain HTML on the server. Place the deep methodology, full chart pack, raw data, or interactive components behind a soft email gate that loads after the crawlable section. The model cites the takeaway. The buyer interested in the methodology submits an email and gets the depth. Both surfaces win.

The technical implementation requires server-side rendering of the head and first section of the page (see Server-side rendering for AI crawlers for the architectural foundation). It also requires careful schema markup so the article body schema reflects the actual public content, not the gated portion. Misrepresenting the available content in schema is grounds for citation devaluation under Google's quality guidelines and is increasingly being penalized by AI model crawlers as well.

The Ungating Playbook: Six Steps Cited With Real Numbers

The operators we worked with through the 200-page test converged on a six-step ungating playbook that delivers most of the citation lift while preserving the highest-intent demand capture.

1. Inventory and stratify your gated assets. Pull a list of every gated asset in your content library — PDFs, downloads, video gates, calculator gates, webinar gates. Stratify them by total form fills over the trailing twelve months and by qualitative content value. Three buckets emerge: high-volume low-quality assets (templates, checklists, generic guides), high-volume high-quality assets (flagship benchmark reports, buyer's guides, frameworks), and low-volume specialty assets (vertical playbooks, archived research). The first bucket should be ungated immediately and replaced with downstream conversion paths. The second bucket goes through the dual-format treatment described below. The third bucket should be ungated and lightly refreshed to recapture citation-worthiness.

2. Convert flagship gated PDFs to dual-format dual-surface assets. For each high-value gated asset, build the HTML article version with the full body content, charts, quotes, and substance. Keep the PDF version as a gated download with a worksheet, methodology appendix, or executive briefing layer added on top of the body content. The HTML wins citations; the PDF captures the executive who specifically wants the handout. Across our test cohort, dual-format pages captured 92 percent of the citation lift of full ungating and 34 percent of original form-fill volume. The combined surface beats either single approach.

3. Engineer the downstream conversion path explicitly. Replace the upstream gate with three downstream conversion points on the HTML article. Position one demo or trial CTA at scroll-depth fifty percent (after the main argument is established). Position the primary CTA at the conclusion of the article. Add a sticky bottom-bar with a soft secondary CTA — newsletter, content alerts, related course — that captures readers who do not convert on demo. The three-CTA structure, modeled on HubSpot Academy and Notion documentation pages, captures 2.1x to 3.4x the conversion rate of a single end-of-article CTA in our test data.

4. Add tracking and citation infrastructure on day one. Before publishing the ungated version, wire up citation tracking via Profound, Otterly, or Peec and tag the page with a unique URL parameter for downstream attribution. Without baseline citation data, the ungating decision becomes unfalsifiable inside the organization — sales will challenge the form-fill loss without seeing the offsetting citation and pipeline lift. The tracking is non-negotiable.

5. Resubmit the ungated assets to LLM crawlers and update LLMS.txt. New URLs are not discovered instantly. Submit the ungated URLs to GPTBot, ClaudeBot, and PerplexityBot via the LLMS.txt manifest at the root of your domain. Verify crawl logs to confirm bot access. Then submit the URLs to traditional search via Google Search Console and Bing IndexNow. Expect first citations within 14 to 30 days; full citation lift typically takes 60 to 90 days as the model retrieval index refreshes.

6. Build a downstream attribution model that captures the new flow. Reorient marketing reporting from same-asset form fills to influenced-pipeline-per-asset and citation-rate-per-asset. The CFO conversation changes from "we lost 16 form fills" to "the asset is now generating $11,900 more influenced pipeline per quarter at 4x the citation rate." Read Dark funnel attribution for the multi-touch model that makes this defensible. The CFO conversation goes badly only when marketing cannot show the offsetting lift in a clean attribution frame.

What Stays Gated: The High-Intent Specialty Layer

Not every asset should be ungated. After running the 200-page test we identified four asset classes where gating preserves both the citation surface and the lead-capture surface because the gated layer is genuinely incremental to the citation-bearing layer.

Raw datasets and methodology appendices. Publish the headline findings, charts, and narrative of original research as a free, citable HTML article. Gate the raw dataset, the full survey response data, the methodology document, and the statistical workings behind an email form. LLMs cite the headline article. Researchers and analysts who want to do their own analysis submit the form and become high-intent contacts. The dataset is genuinely incremental and the citation surface remains intact. This is the structural model behind the original research AEO citation magnet approach Demand Gen Report has documented across B2B publishers.

Interactive calculators with persistent results. Publish a contextual ROI calculator inline on the article — the version that calculates a single output. Gate the persistent results page, the personalized PDF report, and the saved calculator session behind a form. The inline version wins citations and engagement; the saved-results version captures intent.

Live and on-demand product walkthroughs. Embed a short product demo video in the article — three minutes, no gate. Gate the full thirty-minute on-demand product tour behind a form. The short video powers engagement metrics and provides a citable product-evidence layer; the long tour captures genuine buyer interest. Linear, Notion, and Vercel all use variants of this.

Templates that are useful only with the product. Worksheet files, importable templates, and integration-specific assets that are mechanically tied to the product can be gated without citation loss because the article that contextualizes them is the actual citation surface. The template is the consumption artifact; the article is the discovery artifact. Both can win in their own surface.

The unifying principle is that the gated layer must be incremental to the citation layer, not a substitute for it. If gating sits between the model and the substance, the citation surface evaporates. If gating sits between the engaged reader and a deeper consumption artifact, the citation surface survives and the lead capture stays intact.

Progressive Profiling and Newsletter as the Modern Capture Surface

The deepest structural change since 2022 is the migration of lead capture from single-asset form fills to relationship surfaces — newsletter subscriptions, community memberships, content alerts, and progressive profiling across multiple sessions. The reason is buyer behavior. The MarketingProfs 2025 B2B Content Marketing Benchmarks found that the median B2B buyer consumes between 11 and 18 pieces of vendor content before requesting a demo. Asking for an email on the first asset taxes a relationship that has not been built yet.

Newsletter subscriptions in 2026 carry roughly six times the lifetime engagement value of a single asset email capture per the benchmarks we have measured across the same B2B SaaS cohort. The cost of acquisition is lower because the value exchange is clearer (recurring content, not a one-time PDF) and the consent is broader (the reader is opting into a continuing relationship). Newsletter-driven pipeline contribution has overtaken single-asset PDF gates as the primary email capture surface across modern content marketing programs. Substack and beehiiv have made this surface easier to operate than the legacy marketing automation form patterns.

Progressive profiling — capturing one field at a time across multiple sessions rather than a full lead form on the first visit — works the same way. The first session captures an email for content alerts. The second session captures a company name for a personalization upgrade. The third captures a role for a relevance score. By the time the contact is asked for phone or budget information, the relationship has multiple touches and the field-completion rate is much higher than a cold-form attempt would deliver.

The downstream demo CTA, paired with a free read, is the third leg of this capture stack. It works because the buyer has already consumed the substance, evaluated the argument, and self-selected as interested. The demo conversion rate on a downstream CTA after a free read averages 1.8 to 3.2 percent across the test cohort, versus 0.4 to 1.1 percent on a cold landing page demo CTA. The 4x to 6x improvement comes from the qualification work the article does upstream of the click.

Forrester and the Buyer Behavior Shift

Forrester research has documented the buyer journey shift since 2022 and the data informs the gating debate at the strategic level. Forrester's B2B buying research found that 71 percent of B2B buyers begin their research independently before engaging vendors, that the average buying committee touches 27 pieces of content before a demo, and that gated-content forms have declined as a discovery surface since 2022 as buyers route around them via search, AI assistants, and peer communities.

The Forrester data complements the demand-gen mechanics: buyers are deciding without ever filling out a form. They are reading, asking AI, comparing, and self-qualifying. The form is no longer the discovery surface. It is, at best, a hand-raise surface late in the funnel — and even there, the value exchange has to be specific (a demo, a free trial, a calculator output, a custom assessment), not a generic PDF.

The implication for content strategy is that the gate must be moved to the back of the funnel and made specific. The article, the framework, the comparison, the benchmark — these all live in the free, citable layer. The hand-raise — demo, trial, custom assessment, executive briefing — lives in the captured layer. The middle layer is content depth without explicit capture: long articles, embedded calculators, free tools that build relationship rather than extract intent on the first contact.

The strategic move that follows from the Forrester data is to invert the inherited content org chart. Demand-gen teams historically owned gated assets and the marketing pipeline that resulted. The unit economics rewarded volume of form fills. In the AI search era the unit economics reward citation rate and downstream high-intent conversion. The content team should own citation rate as its primary metric and influenced pipeline as its secondary. Form fills become a tertiary diagnostic, not a target.

Operators who have made this org shift — including several profiled in the template downloadable asset playbook — report that the cultural fight inside marketing is harder than the technical implementation. The metric inversion is what unlocks the structural change.

Failure Modes: When Ungating Goes Badly

Ungating does not guarantee citation lift, and three failure modes recur in the data.

The asset was never citation-worthy. Some gated PDFs are gated because the underlying content is thin — a one-page checklist dressed up as a thirty-page eBook with white space and stock imagery. Ungating a thin asset does not produce citation lift because LLMs are not citing the content for substance, they are citing it for retrieval relevance. Fix the substance first, then ungate. The substance fix is usually a 2x-3x expansion of the body content with concrete data, named examples, and quotable claims.

The downstream conversion path was not engineered. Ungating without paired downstream CTAs produces a real citation lift but no demand capture, which is the version of the experiment that gets the project killed by the CRO. The fix is to wire the downstream demo, trial, and progressive-profiling CTAs before the ungating goes live, not after. Treat the conversion infrastructure as part of the ungating release, not a phase two.

The traffic was bot-inflated to begin with. A small share of test pages showed a sharp drop in same-asset form fills that turned out to be bot-driven on the gated version (junk submissions hitting the form). Ungating revealed the inflation. The metric trajectory looked bad on dashboards but the actual high-intent capture was unchanged. Filter for engagement quality, not raw volume.

The 200-page test cohort produced clean lift on 168 of 200 pages, a flat result on 19, and a negative result on 13. The negative cases were concentrated in pages that fit the three failure modes above. Underlying citation lift was consistent in every well-executed case.

Takeaway: Form gates were a strong demand-capture surface from 2010 to 2022. They became an AEO liability the moment LLM crawlers became a primary discovery surface, and the data is now unambiguous. Across 200 pages, ungating delivered a 4.1x LLM citation lift, 2.2x organic traffic gain, and 31 percent net pipeline increase despite a 38 percent decline in same-asset form fills. The right operating playbook is to ungate the substance, build downstream demo and progressive-profiling CTAs, gate only incremental specialty layers (raw data, calculators, long-form tours), and invert the content team's primary metric from form fills to citation rate. Reuters and the Wall Street Journal showed the partial-paywall pattern; HubSpot showed the ungating pattern. The vendors that compound citation share over the next four quarters will be the ones that move first.