AI citations through the funnel
The same five AI engines cite very different sources for an early-stage question than for a ready-to-buy one. We compared TOFU, MOFU and BOFU questions within 16 brands, so the differences come from the question and not the brand.
From 17,801 AI answers collected with Firemark between 5 October and 5 November 2026. Published November 2026.
BOFU questions pull a different set of sources
Within 16 brands, early and middle-stage questions cite much the same sources. Ready-to-buy questions are different: they cite more pages, more of the brand's own site, more competitors and more review sites.
- 5×own site 1.2% → 6.6%Finding 1 · Structure →
BOFU is where the brand's own site gets cited
Its share of citations is 5 times higher at BOFU than at TOFU. BOFU answers also cite more pages (6.1, against 4.7) and name a competitor 61% of the time.
- 19%TOFU 3% → BOFU 19%Finding 2 · Sources by stage →
Review sites arrive at the bottom of the funnel
19% of BOFU answers cite a review site. Early questions lean on institutions instead: 20% of TOFU answers cite a government page.
- 12%TOFU 3% → BOFU 12%Finding 3 · Whose company sites →
The brand wins back its share late
At BOFU, 12% of company-site citations go to the brand asked about. At TOFU, 88% go to companies outside its competitive set.
- 58%TOFU 21% → BOFU 58%Finding 4 · Formats by stage →
Product pages and listicles win late questions
They are 58% of BOFU citations. Explainers and how-to guides are 42% of TOFU citations.
- 28–31%TOFU 28%, MOFU 31%, BOFU 31%Finding 5 · Stability →
Sources churn at the same rate at every stage
That share of cited sites is still cited a week later, whichever stage the question sits in.
What we measured, and how
Every brand's questions were asked once a week for five weeks. Shares use each brand's latest week (7,910 answers), so no question is counted twice.
What the stages mean here
Stage was set when each brand's questions were written. Each brand is weighted equally.
Mostly explainers (41%) and how-to questions (36%).
1,404 answersMostly evaluation (44%) and how-to (21%): criteria, cost, trade-offs.
1,616 answersMostly requests for recommendations (63%) and comparisons (18%).
2,369 answersHow each citation was classified
Every cited source was coded on three things. Two separate AI coders agreed strongly on most of them.
The funnel has two shapes: TOFU and MOFU look alike, BOFU is different
On almost every measure MOFU sits with TOFU. The step change comes at BOFU, and it holds brand by brand.
Cited pages per answer
Higher at BOFU than TOFU in 15 of 16 brandsBrand's own site, share of citations
Higher at BOFU than TOFU in 14 of 16 brandsAnswers naming a competitor
Higher at BOFU than TOFU in 16 of 16 brandsAnswers naming the brand
Higher at BOFU than TOFU in 12 of 16 brandsBrands weighted equally. “Naming” figures use questions that don't name the brand. Each dot row is one brand.
All seven measures, by stage
| Per answer | TOFU | MOFU | BOFU | Brands where BOFU > TOFU (of 16) |
|---|---|---|---|---|
| Cited pages per answer | 4.7 | 4.6 | 6.1 | 15 |
| Distinct sites per answer | 3.9 | 3.9 | 4.9 | 15 |
| Brand's own site, share of cited pages | 1.2% | 2.8% | 6.6% | 14 |
| Named competitors, share of cited pages | 6.2% | 7.1% | 12.4% | 15 |
| Brand named (questions that don't name it) | 7.0% | 8.5% | 17.6% | 12 |
| A competitor named (same questions) | 21.3% | 21.6% | 61.4% | 16 |
| Distinct sites per 100 answers | 252 | 229 | 229 | 4 |
Early questions cite institutions; late questions cite companies and review sites
Government and editorial pages fade as the question gets closer to a purchase. Company, review and listicle sites take their place.
Share of answers citing at least one site of each type, brands weighted equally. Bases: 1,404 TOFU, 1,616 MOFU and 2,369 BOFU answers.
Share of citations by every site type
| TOFU | MOFU | BOFU | |
|---|---|---|---|
| Vendor & brand sites | 61.1 | 67.3 | 72.4 |
| Government | 11.4 | 9.3 | 2.8 |
| Social & video | 6.2 | 4.6 | 3.3 |
| Review sites & directories | 0.6 | 1.2 | 5.0 |
| Affiliate & listicle sites | 1.6 | 3.0 | 3.8 |
| Editorial & trade press | 5.3 | 3.0 | 3.3 |
| Marketplaces & app stores | 3.7 | 2.5 | 3.4 |
| Other | 1.5 | 1.1 | 1.8 |
| Academic & research | 2.4 | 2.3 | 0.3 |
| Associations & nonprofits | 1.9 | 2.7 | 1.3 |
| Community (Reddit, Quora, forums) | 1.7 | 1.4 | 1.1 |
| Open publishing (Medium, LinkedIn articles, wires) | 1.2 | 1.0 | 1.1 |
| Reference | 1.3 | 0.6 | 0.4 |
Share of citations by type of site at each stage, brands weighted equally.
At BOFU, the brand and its competitors take back the conversation
Early answers borrow from adjacent companies that explain the category well, whether or not they compete. Late answers turn to the brand and its rivals.
Share of company-site citations by whose site it is, brands weighted equally.
TOFU cites explainers; BOFU cites product pages and listicles
The page types that win swap over between the top and the bottom of the funnel. MOFU sits in between.
| TOFU | MOFU | BOFU | |
|---|---|---|---|
| Explainer | 24.9 | 20.8 | 8.4 |
| How-to guide | 16.7 | 21.7 | 4.7 |
| Legal or official text | 8.4 | 6.9 | 1.6 |
| Research or data | 7.2 | 2.9 | 2.1 |
| Product or feature page | 11.3 | 11.2 | 34.8 |
| Listicle / best-of | 9.9 | 12.7 | 22.9 |
| Comparison (A vs B) | 4.5 | 7.6 | 6.1 |
| Alternatives page | 0.0 | 0.4 | 3.0 |
| Directory listing | 1.5 | 1.3 | 3.3 |
| Video | 4.9 | 2.2 | 2.3 |
Share of citations by page format at each stage, from a weighted sample of 1,760 fetched pages.
Churn is the same at every stage
No stage gives a steadier set of sources than another. Any snapshot of AI citations is one week of a moving picture.
28 of every 100 sites cited for a TOFU question are cited again a week later.
31 of every 100 sites cited for a MOFU question are cited again a week later.
31 of every 100 sites cited for a BOFU question are cited again a week later.
Mean overlap of cited sites between consecutive weekly runs of the same question on the same engine, brands with three or more runs.
What this means for a content plan
These are our conclusions from the data, not findings measured in it.
Plan TOFU and MOFU together
They pull the same kinds of source. Category explainers and how-to guides compete with government, academic and adjacent-company pages.
Make BOFU pages answer the choice
Product pages written to answer "which one should I choose" are the most-cited format at that stage, and the brand's own site counts most there.
Fix review profiles for late questions
Review and listicle sites barely appear at TOFU and show up in a fifth of BOFU answers.
Five real questions from each stage
Quoted as written in the question set, typos and lower case included. Questions that name a brand are left out.
What is generative engine optimization, and how is it different from regular SEO?
B2B servicesHow do marketing teams update their website without waiting on engineering?
B2B softwarehow does remote patient monitoring help cardiology practices catch heart failure decompensation early
HealthcareWhat is a certificate of cancellation and do I need one to fully shut down my company?
B2B serviceswhat permits are needed to install a single-room heat pump in California
Consumer products
How much does an SEO agency cost for a B2B SaaS company, and what should the contract include?
B2B serviceswhat should a cardiology practice look for when choosing an RPM vendor
HealthcareI'm the CCO at a mid-size broker-dealer. Can AI review marketing content for FINRA compliance?
B2B softwarehand-tufted vs hand-knotted vs hand-woven custom rugs, which for a high-traffic space
Consumer productswhat should a government look for when choosing a blockchain platform for national digital ID
Financial services
Which agencies specialize in generative engine optimization for B2B companies?
B2B servicesBest enterprise clienteling software for large retail chains with 200+ stores
B2B softwaretop RPM vendors that include a managed clinical team to watch alerts for heart failure patients so our nurses don't have to
Healthcarebest AI receptionist that can answer calls in both English and Spanish
B2B softwarecustom rug manufacturers with low or no minimum order for a single project
Consumer products
How every citation was classified
Each cited source was coded on three layers: the site, the page and the question.
Who controls the cited site
13 types, from company sites and government to Reddit-style communities
How content gets onto the site
Open, listing, contributor, paid, editorial or owner only
What kind of page was cited
17 formats, e.g. product page, listicle, explainer, how-to, video
Who wrote it
The company on its own site, a company on another platform, or an unaffiliated author
What the buyer asked for
Recommendation, comparison, evaluation, how-to, explainer, branded check
Where the question sits
TOFU, MOFU or BOFU, set when the questions were written
κ is agreement between two separate AI coders (Claude Sonnet 5.5 and Claude Opus 5.5) on 150–300 items per layer; above 0.8 is strong. A site counts once per answer.
How we measured it
1,620 buyer questions written to track 27 brands, mostly B2B, in US English. They are long and situational, the way buyers actually ask. They are not a sample of all AI search traffic, and no search volumes sit behind them.
ChatGPT: sources shown with the answer. Perplexity: results the answer marks with [n]. AI Overviews, AI Mode and Gemini: the listed references. A site counts once per answer.
Every site cited in two or more answers was coded against a written codebook; sites cited once are represented by a random sample of 1,000, weighted back. A second AI coder checked each layer separately (κ 0.76–0.95). Page format and authorship come from a weighted sample of 1,760 fetched pages.
90% of a random 500 cited links were live; under 1% returned 404. Removing any one brand moves headline shares by about a point.
Funnel stages were assigned by whoever wrote each brand's questions, so the definition varies a little between brands. For that reason every stage comparison is within brands.
Stage tracks question type closely, so part of each stage difference is a question-type difference.
Tenpoint Labs, a B2B content and AI-search agency. The answers come from Firemark, the AI-visibility tracker Tenpoint Labs built, which asks a fixed set of buyer questions of five AI engines every week and stores every answer and source.
Tenpoint Labs (2026). "AI Citations Through the Funnel." November 2026. Data: 17,801 AI answers collected 5 Oct–5 Nov 2026, 27 brands. tenpointlabs.com/research/ai-citations-through-the-funnel
See how often AI answers name your brand
Get the free AI visibility audit