Why 55% of all AI citations happen at the evaluation stage, and what that means for how you spend your content budget.
Last updated: June 4, 2026
Key takeaways
- HubSpot's 2026 study of 10,000 URLs across 8 AI engines found that 55% of AI citations happen at the evaluation stage of the buyer journey, not the awareness stage.
- The funnel for AI engines is upside-down compared to Google search. Most clicks from Google come from top-funnel awareness queries. Most citations from AI engines come from bottom-funnel buying queries.
- If you're investing your content budget in "what is X" educational content, you're optimising for 12% of the citation pie while ignoring the 55% slice.
- The right move is to produce comparison and evaluation content for the prompts your buyers type when they're choosing, not when they're learning.
- For trademark owners and brand operators, that means writing brand-vs-brand comparison pages, evaluation scorecards, and "best X for Y" content with concrete numbers and named examples.
What does "bottom of the funnel" mean for AI search?
The buyer journey is usually drawn as a funnel with four stages. HubSpot's own framing calls them awareness, consideration, evaluation, and decision. Other frameworks use slightly different names, but the underlying idea is the same: buyers start broad and narrow toward a choice.
For traditional search engines, the top of the funnel is where the volume is. Most Google searches are exploratory. "What is a trademark", "how do trademarks work", "what is the USPTO" each get tens of thousands of monthly searches. The bottom of the funnel, the "should I file a trademark for my Shopify store" queries, gets a fraction of that volume.
That ratio reverses for AI engines. HubSpot's 2026 AEO research, which analysed 10,000 URLs across AI Overviews, Gemini, Perplexity, ChatGPT, SearchGPT, Grok, Google AI Mode, and Copilot, found that AI citations are heavily skewed toward the bottom of the funnel:
- Solution education (awareness): 12% of citations
- Problem exploration (consideration): 14% of citations
- Solution comparison (evaluation): 19% of citations
- Solution evaluation (decision): 55% of citations
More than half of every AI engine's citation activity happens when the buyer is choosing. Three quarters of it happens at or below the evaluation stage. This is the opposite of how Google traffic distributes.
Why do AI engines skew so heavily to the bottom of the funnel?
Two reasons.
First, AI engines are used differently than search engines. People type "best CRM for a 12-person agency" into ChatGPT because the AI can reason about their context, weigh tradeoffs, and produce a synthesised recommendation. They don't type "what is a CRM" into ChatGPT because the AI's training data already gave them a baseline understanding. Top-funnel education has largely been outsourced to baseline AI knowledge. The interesting prompts are the ones where the buyer needs help with a decision.
Second, AI engines are better at the bottom of the funnel. A ranked list of ten blue links is a poor format for "best X for my specific situation". A synthesised paragraph that names three options and explains why is a great format. Buyers learned this quickly. Within twelve months of ChatGPT launching, evaluation-stage prompts moved from being a curiosity to being the default.
What does a bottom-funnel prompt actually look like?
Specific. Named. Constrained. Three rough patterns:
- "Best X for Y" prompts. "Best trademark filing service for a US-based SaaS startup", "best brand monitoring tool for a portfolio of 5 brands", "best NICE class to file in for an AI infrastructure product".
- "X vs Y" prompts. "LegalZoom vs Trademark Engine for first-time filing", "filing in Class 9 vs Class 42 for a SaaS product", "Trademark Now vs Filing 1A vs going direct to a lawyer".
- "Should I do X" prompts. "Should I file a trademark before I incorporate my company", "should I trademark a phrase that's been used in our marketing for two years", "is it worth filing internationally if I'm only selling in the US right now".
Compare these to top-funnel awareness queries: "what is a trademark", "how do trademarks work", "what is the USPTO". The top-funnel queries are about understanding. The bottom-funnel queries are about choosing. The buyer who types a bottom-funnel prompt has a problem, has narrowed their options, and wants a recommendation.
How much should I be willing to spend per bottom-funnel citation?
This depends on your unit economics, but here's a useful exercise. If your average customer is worth $4,800 over their lifetime, and 5% of buyers in your category use ChatGPT as part of their evaluation, and your close rate is 20% of evaluation-stage buyers you reach, then a single citation in a high-intent prompt that gets typed 500 times a month is worth roughly:
500 prompts/month × 5% AI-using buyers × 20% close rate × $4,800 LTV = $24,000 per month per cited prompt
That math is illustrative, not exact. But it makes the point. The economics of bottom-funnel AI visibility are not subtle. A single high-intent prompt where you're cited is worth more than a mid-six-figure annual SEO investment in top-funnel content that AI Overviews now eat for free.
What content actually wins at the bottom of the funnel?
The HubSpot study identified the structural signals that produced the biggest measurable lifts in AI citations. None of them involve link-building. They're all on-page and structural:
- Question-style H2 and H3 headings: +28% citation lift on AI Overviews, +19% on Gemini.
- FAQ schema (JSON-LD): +24% on AI Overviews.
- Twenty or more outbound links to authoritative sources: +19% on AI Overviews.
- Visible on-page FAQ section: +18%.
- Visible "last updated" date near the headline: +8% AIO, +6% Perplexity, +5% Gemini.
Apply this checklist to a comparison page, an evaluation scorecard, or a "best X for Y" piece written for a real bottom-funnel buyer prompt, and you stack the deck in your favour. The piece reads well to humans and it carries the structural signals AI engines weight heavily.
This is also why Content Marketing Institute, Search Engine Journal, and other trade publishers have been quietly retooling their editorial briefs in 2025 and 2026. The format that works for AI citations is the same format that works for human readers who are about to buy: question-framed, evidence-driven, structured for scanning, freshly updated.
What does this mean for trademark owners and brand operators specifically?
If you operate a brand or hold a portfolio of trademarks, the bottom-funnel content opportunity sits at the intersection of brand strategy and trademark filings:
- Brand-vs-brand comparison pages. Compare your brand against your three top competitors on real, verifiable signals. Trademark filings (via the USPTO TSDR database or our own Compare strategies tool), NICE class coverage, expansion velocity, TTAB activity.
- Category evaluation guides. "How to choose a [category] vendor in 2026" written with concrete criteria, named brands, and links to valuable brand profiles on WikiTrademarks.
- "Best NICE class for X" pages. "What NICE class should I file in for a SaaS product targeting legal professionals" with examples from real filings.
- TTAB analysis pieces. Specific TTAB proceeding teardowns with proceeding numbers, decisions, and what they mean for similar disputes.
Each of these maps to a real bottom-funnel buyer prompt. Each builds on proprietary data the AI cannot generate from scratch. Each gives the AI a reason to cite you as the authoritative source rather than summarising someone else's content.
How do I know whether I'm actually winning bottom-funnel citations?
You measure it. Pick the 10 to 20 bottom-funnel prompts your buyers actually type, run them through ChatGPT and the other engines weekly, and log what you see. Without measurement, you're guessing whether your content is working. With measurement, you find out within four weeks whether a new comparison page started showing up in AI answers or got ignored.
If you want this measurement automated against real chatgpt.com data (which differs meaningfully from the OpenAI API), our AEO Tracker handles the weekly capture and gives you a structured report with diffs. It's in closed beta with a waitlist.
Frequently asked questions
Why is bottom-of-funnel content more valuable than top-of-funnel content for AI citations?
Because the citations happen there. HubSpot's 2026 research found 55% of AI citations happen at the evaluation stage and only 12% at the awareness stage. A buyer who types an evaluation prompt is closer to a purchase decision, so each citation is worth meaningfully more in expected revenue.
What's an example of a high-value bottom-funnel prompt?
Something specific, named, and decision-oriented. "Best trademark filing service for a US-based SaaS startup", "Class 9 vs Class 42 for a SaaS product", or "LegalZoom vs filing direct for a clothing brand". The buyer typing those prompts is choosing, not exploring.
Does this mean I should stop writing top-of-funnel educational content?
Not entirely, but you should rebalance. Top-of-funnel content still helps with brand recall and links from references. It just no longer drives the traffic or AI citations it used to. The HubSpot data shows informational queries lose 25 to 35 percent of clicks to AI Overviews. Treat top-funnel as supporting scaffolding, not as a standalone traffic strategy.
How do I produce comparison content if I don't have direct competitive intel?
Use public sources. For trademark and brand strategy comparisons, USPTO filings are open data, and tools like our Compare Strategies tool let you compare two brand owners' filing patterns directly. For SaaS comparisons, public pricing pages, G2 reviews, and direct testing get you most of the way there.
Will AI engines still cite my content if I don't have lots of backlinks?
Yes. HubSpot found backlink count has near-zero correlation with AI citations. What matters is on-page structure (question H2s, FAQ schema, visible "last updated" date), original data, and presence on the third-party surfaces AI engines pull from (Reddit, LinkedIn, YouTube, trade press).
How long should a bottom-funnel article be?
Long enough to actually answer the buyer's question and short enough to respect their time. For evaluation guides, 1,500 to 2,500 words is typical. The structural signals matter more than the word count: clear question H2s, statistics, named examples, visible FAQ, and a "last updated" line.
Want to know whether ChatGPT mentions your brand in the bottom-funnel prompts your buyers actually type? Join the AEO Tracker waitlist. We'll be in touch when an onboarding slot opens.