Vendor discovery01
84%3.5× in 24 months
CMOs using AI to find vendors
202424%
202684%
In two years it went from a minority habit to the default first move for the person who signs your contract.
Our process · The GEO Stack
Get in the answer. Get the pipeline.
Your buyers ask an AI assistant who to shortlist before they ever reach your site. We make you the answer, on a published schedule, and prove what it's worth in your CRM.
✓First visibility in 2 to 4 weeks✓Pipeline, not traffic✓Founder-led
The shape of the engagement
Three layers · Twelve weeks · One scoreboard
Attribution wired and reportedWeek 1 on
01 · Solution-aware pagesWeek 2 on
02 · Citation buildingWeek 3 on
03 · Problem-aware bridgeWeek 8 on
Tracked across all major LLMs
01
What changedAI search buyers arrive further down the funnel, and they buy faster.
Google used to do two jobs. Now it does one. Discovery moved into the prompt, and verification stayed on Google, which means the shortlist is written before anyone lands on your site.
The old journey
Google searchBroad, exploratory · 3-5 words
"invoicing software"
- Scans ten blue links
- Opens four to six tabs
- Pieces together their own shortlist
- Enters your site to browse
Arrives cold, comparing from scratch. Verification only, and only once the name is already known.
The new journey
ChatGPT · Gemini · PerplexitySpecific, solution-aware · 15-30 words
"Best invoicing software for a 20-person agency with recurring billing"
- Asks a sophisticated question upfront
- Refines with two to four follow-ups
- Receives a ranked recommendation
- Enters your site to validate a decision
Arrives pre-qualified, already shortlisted. This is where the decision gets made, in a conversation you weren't part of.
Where the journey starts02
68%Begin in a prompt
Start in AI, not a search engine
Search first32%
AI first68%
Two thirds of buying journeys now begin somewhere your rank tracker cannot see, and end with a name already chosen.
Quality of the visit03
3-4×Conversion vs. organic
How AI referrals convert
Organic1×
AI referral3.5×
Tiny in volume, the highest intent in your funnel. They arrive to validate a recommendation, not to browse.
Source · Wynter, Jan 2026, n=101 B2B SaaS CMOs · Conversion rate from client funnels
02
How we do itThe GEO Stack. Three layers, built in order.
Twelve months of data across 20+ B2B SaaS companies. Sequencing matters. Skip a layer and the next one underperforms.
01
Solution-aware pages, bottom of funnel first
The content AI pulls from when someone asks which product to buy. Not "what is customer success software", which the model answers itself and you never get seen. Instead: best-for pages, comparisons, alternatives and vertical pages, written so a model can quote them cleanly.
1 to 4 weeks
Time to first visibility
02
Citation building on the sources models trust
Presence gets you into the conversation. Frequency makes the model confident recommending you over a competitor. We extract the domains AI already cites for your prompts, find where you're absent, and earn mentions there. Mention-building, not link-buying.
4 to 12 weeks
Compounding window
03
The problem-aware bridge
A VP of Finance types "how do I reduce invoice errors across three entities". The model bridges that pain to a software category. If your content is the source it pulls from, you enter at the exact moment of realization. Highest intent in the stack, which is why it comes last.
Week 8 on
Full-funnel coverage
Two things move the needle: bottom-of-funnel presence and citation building. Everything else is a rounding error until those are in place.
Schema tweaks, llms.txt, embedding tricks, prompt hacks, AI sitemaps. We've tested the full surface across 20+ campaigns. They're real, and they're secondary. We do the boring work first, in public, on a schedule.
Schemallms.txtFreshness signalsAI sitemapsTechnical SEO
03
What happens, week by weekThe 90-day plan, on one page.
Four waves, sequenced to recover fast and then compound. You get this exact plan, dated, before anything ships.
The 90-Day AI Search Program · standard build
Run by usNeeds you
WorkstreamW1W2W3W4W5W6W7W8W9W10W11W12
Wave 0 · Set up
Weeks 1-2
Kickoff and accessCRM, analytics, SMEs
Attribution wiringTwo-Witness Rule, live first
Audit and prompt mapDiagnosis, kill-tested
Build · the three layers
Weeks 2-12
Wave 01 · RecoverLayer 01 · BOFU pages
Wave 02 · CompareLayer 01 · Switcher capture
Wave 03 · CiteLayer 02 · Mention frequency
Wave 04 · CompoundLayer 03 · Problem-aware bridge
Always on · measurement
Weeks 1-12
Visibility tracking5 engines, weekly re-runs
Reporting cadenceMonthly board + strategy call
Milestones
First AI visibility
First citations land
First attributed demos
Day-90 scoreboard vs. forecast
Wave 01Weeks 1-12 · Ongoing
Recover
Your commercial terms on dedicated, citable pages. This never stops: the solution set expands every week for the full 90 days.
- →Technical audit and fixes
- →Prompt map, kill-tested
- →Solution pages, weekly
- →CRM attribution live in week 1
Wave 02Weeks 3-6
Compare
Own the head-to-heads so the model quotes your comparison instead of a competitor's or a review site's.
- →Alternatives pages
- →VS pages, made citable
- →Pricing and persona pages
- →Review velocity started
Wave 03Weeks 3-12 · Ongoing
Citation building
Get onto the sources the models already quote, then keep landing on them. Frequency is what turns presence into a recommendation.
- →Review and roundup presence
- →Industry media placements
- →Community and expert commentary
- →First attributed demos
Wave 04Weeks 8-12
Compound
Bridge to the pain buyers describe before they know the category name, then grade the quarter against the forecast we set at kickoff.
- →Problem-aware bridge pages
- →Refresh and re-optimize wave 01
- →Day-90 scoreboard
- →Next-quarter plan
Every week
Artifacts land in your Slack: pages, briefs, placements, and the prompt runs where you were and weren't named.
Every month
One scoreboard: demos, pipeline, citations and rankings, reconciled by the Two-Witness Rule and graded against the forecast.
Every quarter
Re-cut the model against real numbers, re-prioritize the roadmap, and agree the next 90 days.
Your side
Under two hours a month. Approvals, occasional SME input, one strategy call. If we need more, we designed it wrong.
04
How we prove pipelineThe Two-Witness Rule.
Analytics has exactly one witness to where a lead came from, and AI assistants blind it. So we add a second witness, the buyer, and rule on the pair.
What it is: every contact gets read against two independent witnesses instead of one. Witness one is the machine, the referring URL your CRM records. Witness two is the human, what the buyer typed into "How did you hear about us?" A channel is credited on the evidence of the pair, under a written rule, not on whichever number is more flattering.
Why we built it: the machine went blind. AI assistants strip the referrer, so a buyer who spent a week in ChatGPT lands as "Direct." We pulled 8,307 contacts from one growth-stage funnel and read them both ways. The CRM found 11 AI-sourced leads. The buyers themselves named an AI assistant 809 times.
Witness one · CRM referrer forensicsMachine · Blind to AI
Witness two · "How did you hear about us?"Human · Biased by recall
Credit assigned by written rule, ranges not point estimatesAuditable
Neither witness is the truth on its own. One cannot see AI at all; the other remembers imperfectly. The rule is what turns two unreliable witnesses into a number your CFO can actually defend.
One funnel · 8,307 contacts
The same leads, read two ways
11
Logged by the CRM
as AI-sourced
809
Buyers who named an
AI assistant themselves
A 73× gap on the channel the CMO was being asked to fund
How we classify every contact
Witness oneCRM referrer · sees clicks, blind to AI
Witness two"How did you hear about us?" · sees intent, imperfect recall
ConfirmedHard count
CRM saysOrganic search
Buyer says"Google"
Both witnesses point at the same channel, so the lead is counted as-is. No correction, no judgment call.
RecoveredCredited to AI
CRM saysDirect
Buyer says"ChatGPT"
The machine is structurally blind here: AI assistants strip the referrer. The buyer is the only witness who can see it, so the buyer wins.
Single signalProbable
CRM saysA named channel
Buyer saysBlank
Only one witness spoke. We take the signal we have, label it probable, and keep it out of the hard count.
ConflictResolved by rule
CRM saysPaid search
Buyer says"Perplexity"
Both witnesses could see something, and they disagree. Whoever could actually see it wins: the machine takes paid clicks and branded direct, the human takes demand creation.
DarkNamed, never guessed
CRM saysDirect
Buyer saysBlank
Neither witness can name a source. It stays in a bucket labelled exactly that, and it is reported at its real size rather than allocated to a channel that looks good.
Every contact lands in exactly one case. Corrections ship as a range with a stated base case, and raw always appears next to corrected.
The conflict rule
Whoever could actually see it, wins
The human wins on demand creation: AI search, communities, word of mouth. A referrer-based classifier structurally cannot see those. The machine wins on demand capture: paid clicks and branded direct, where the click record beats recall.
The honesty rules
Ranges, not false precision
Not everyone answers the form, so corrections ship as a band with a stated base case, never a single number. Raw and corrected always appear as a pair. And the leads nobody can see stay in a bucket labelled exactly that.
Why this matters for your budget: the standard advice, only scale what every source agrees on, structurally defunds AI search, because AI is the one channel where the witnesses almost never agree. Follow it faithfully and you cut the channel that's actually growing.
05
What comes out the other sideVisibility first. Then pipeline.
Three programs, three categories, the same three layers. Numbers from the monthly reports we send anyway.
RasaConversational AI
AI search became a demand channel in one quarter
Organic sessions, AI referrals, category presence and attributed demand all moved together one quarter after launch.
+127%
Attributed demo requests, search and AI
WaitWellQueue management
Four months in, running ahead of forecasted demos
Nearly half the demo pipeline was filed as Direct. The Two-Witness Rule recovered it, and the page program kept feeding it.
2.9×
Demos vs the Realistic forecast published pre-kickoff
LendbuzzAuto finance
AI search and organic became the growth engine
A page set that did not exist twelve months ago now carries the growth and is the most-cited name in the category’s AI answers.
+200%
Core conversion event #1, from organic, year over year
Straight answers
Asked on every call.
If yours isn't here, bring it. Same unpolished answer either way.
01How is this different from SEO?+
It isn't separate from it. The same pages that earn rankings are the pages models quote, and the mentions that make a model confident are the mentions that build authority in Google. We run one program across both surfaces because your buyers move between them in a single afternoon.
02How fast does anything show up?+
First visibility on the target prompts typically lands two to four weeks after the first pages ship. Citations compound from week four. Attributed demos usually start inside the first 90 days. Every one of those dates is on the plan above before we start, and the day-90 scoreboard grades us against it.
03Can you really attribute AI search?+
Not to per-lead certainty, and anyone who claims that is selling you a dashboard. We produce corroborated ranges and trends from two witnesses, with the unknowns named openly. That's a picture a CFO can act on, and one they can't dismantle in a single question.
04What do we have to do?+
Under two hours a month: approvals, occasional SME input, and one strategy call. Kickoff needs a deeper session and access to your CRM and analytics. If we're asking for more of your time than that, we designed it wrong.
05Who is this wrong for?+
Teams that need leads next week, have no working sales motion, or can't give a channel 90 days. This compounds. It rewards B2B SaaS teams selling considered deals who want a durable channel rather than a spike.
Next step
Your GEO plan, built on your numbers.
Give us your domain and your pipeline goal. We map the prompts your buyers are asking, build your GEO marketing plan against them, and forecast the pipeline the next 90 days can realistically produce.
Your plan · Walked through live
Founder-built
Justin Berg
Founder · Rock the Rankings
What we walk through
01Your prompt map: the buying prompts and queries that land you on a vendor shortlist.
02Where you show up today vs. your three closest competitors and where the gaps exist.
03The 90-day sequence: what we would ship, in what order.
04The 3-scenario ROI forecast, run on your actual numbers.