The method
The Sourcelift Loop, shown, not described
The Sourcelift Loop is our five-step method for getting brands cited by AI engines: a Baseline on day 0, then Fix, Ship, Earn and Prove every month. Results are measured on the same locked set of buying prompts, across five AI engines.
Read the film as text
- Day 0. A customer asks ChatGPT for the best running shoes for flat feet under $150. Three competitors are named, your brand is not. Ask again and the list changes, so one screenshot proves nothing.
- Baseline. 32 real buying prompts, asked on five engines: 160 answers, locked on day 0. Your brand is cited in 17 of them, 11%.
- Fix. Firewall rules were blocking two AI search bots, OAI-SearchBot and Claude-SearchBot; now they can read the site. Product schema goes on 48 product pages and one brand description across 9 profiles.
- Ship. Four answer-first pages in the month, each one aimed at prompts in the set, starting with a tested guide to the best running shoes for flat feet.
- Earn. Listings on the pages engines already cite for the category: an independent review roundup and a running magazine buyer’s guide, with two more in talks. No link spam.
- Prove. Day 30, the same 32 prompts run again: 14 new citations and 2 lost, every change counted. 29 of 160.
- Months 2 and 3. What Prove shows sets the next Fix, Ship and Earn: 44 of 160, then 61.
- Day 90 report. AI visibility at 38%, up 27 points from 11%, with the citation rate per engine and against named competitors. Then the same question from day 0, asked three times on ChatGPT: your brand is cited every time, and 4 of the 5 engines now cite it.
01 / Step by step
What happens in each step.
- 01
Baseline
Day 0What we do
We lock 20 to 40 real buying prompts for your category and measure where you and your competitors stand today, on five engines.
What you get
The day 0 report: your prompt set and where you and your competitors stand, engine by engine.
- 02
Fix
Every monthWhat we do
Entity and Technical GEO: one consistent story about who you are everywhere engines look, plus schema, architecture and crawl access for AI bots.
What you get
Schema, entity and crawl fixes, shipped on your site and on the profiles engines read.
- 03
Ship
Every monthWhat we do
Answer-first content every month: comparison pages, commercial pages, citable guides and FAQs, built to be the source engines quote.
What you get
A fixed number of answer-first pages every month, each one aimed at prompts in your set.
- 04
Earn
Every monthWhat we do
Mentions in the comparison pages, lists, directories and communities AI engines already trust. We do not buy link spam.
What you get
New mentions on the sources engines already cite, logged in the monthly report.
- 05
Prove
Every 30 daysWhat we do
The same locked set runs again. You see citations won, share of voice and the delta per engine, and the next loop starts from there.
What you get
The monthly report: citations won, share of voice and the delta per engine, with the raw data.
Repeated every month: what Prove shows sets the priorities for the next Fix, Ship and Earn.
02 / The honest problem
GEO has a measurement problem. Solving it is the product.
The problem
Citations appear and disappear. Models get updated. The same question asked twice can name different brands. This is why most agencies describe their method in adjectives, and why so many clients leave after a few months: nobody can show them what changed.
Our answer
It is boring and it works: lock the instrument, then measure relentlessly. Same prompts, same engines, same scoring, every 30 days. When the delta is positive you know it. When it is not, you know that too, and so do we. That symmetry is what makes the number trustworthy.
03 / The instrument
What a monthly report looks like.
Baseline and Prove run on the same instrument, the Sourcelift Visibility Baseline: the same prompts, the same engines and the same scoring, every time.
AI visibility
38%
▲ 27 ptsvs 11% on day 0
Cited answers
61/160
+20 new · −3 lost vs day 60
Share of voice
46%
#1 of 3 brands · #3 on day 0
Mentioned, no link
18
The next ones to turn into citations
Citation rate vs competitors
% of answers that cite each brand with a link
Share of voice
Your share of the citations the three brands split
- 1 Your brand 46%
- 2 Competitor A 36%
- 3 Competitor B 19%
From #3 to #1 in 90 days
All 160 answers, every run
32 buying prompts × 5 engines. Locked on day 0.
- Cited with a link 61
- Mentioned, no link 18
- Absent 81
Biggest movers
Engines citing you, day 0 → now
- 10 do flat feet need stability shoes 0 → 5
- 01 best running shoes for flat feet under $150 0 → 4
- 12 best running shoes for heavier runners 0 → 4
- 19 best shoes for nurses who also run 0 → 4
- 03 best running shoes for overpronation 2 → 5
AI crawlers
Visits in the 7 days before each run
- OAI-SearchBot Search 196
- GPTBot Training 118
- PerplexityBot Search 117
- ChatGPT-User User fetch 84
- Claude-SearchBot Search 81
- ClaudeBot Training 70
- Perplexity-User User fetch 52
The prompt set
20 to 40 real buying questions in four types: category ("best X for Y"), comparison, alternatives and use case. Locked at day 0.
The engines
ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews. Five surfaces, measured separately, because they behave differently.
The metrics
Citations with a link, per engine. Status per prompt: recommended, mentioned or absent. Share of voice against the competitors that appear.
The cadence
Baseline at day 0, then a new measurement every 30 days. Deltas against day 0 and against last month, always.
04 / The sprint
How a sprint runs, week by week.
The optional sprint is a fixed-price start: one baseline, twelve weeks of work and one closing report. Growth programs, with or without a sprint first, run the Loop every month and report every 30 days.
Baseline
Prompt set defined with you, first full measurement, competitor map, entity and technical review. You get the day 0 report.
Fix, Ship and Earn, in impact order
Highest commercial value first: entity clarity, then answer-first pages built around your prompts, then structured data, then external citations.
Prove
The full set runs again. The closing report shows the delta per engine and what to prioritize next. This report is what turns a sprint into a growth program, or into a clean goodbye.
The objection this page exists to answer: "AI visibility cannot be measured." It can, just not casually. See how we treat proof, or check what the sprint costs.
05 / FAQ
Measurement, common questions
How we measure, what we commit to and what we will not claim.
Is running the prompts the service?
No. The prompt set is how we measure. The service is what happens between measurements: content shipped, Technical GEO fixes, entity updates, citation outreach. The locked set exists so you can verify that the work moved the number.
Why does the prompt set stay locked?
Because a moving instrument cannot prove anything. If you change the questions every month, you can manufacture any trend you like. We freeze the set at day 0 so every re-measurement compares like with like. New prompts can be added as a separate, clearly labeled set.
AI answers change between runs. How is that measurable?
Single answers vary; distributions do not. We run the full set on a fixed cadence and score presence, position and citations across all of it. One screenshot proves nothing, which is exactly why we never sell screenshots.
How do you know the movement came from your work and not a model update?
We cannot isolate every factor, and we will not pretend to. What the locked set gives you is a fair comparison: the same prompts, run the same way, with your competitors in the same answers. A model update that reshuffles sources moves their numbers too, not only yours. When your share of voice climbs while theirs holds, that is the work.
Can you guarantee first position in ChatGPT?
No, and nobody honestly can. No agency controls what a model outputs, which is why we sell measured work, not rankings. What we commit to is the Sourcelift Loop itself: the prompts we lock, the work we ship every month and a report with the change on each engine, flat months included.
Does AI visibility turn into sales?
Partly in ways you can see, partly not. When an engine links to you, the visit arrives as referral traffic from that assistant. When it only names you, the buyer often searches your brand later, and no tool attributes that cleanly. So we treat citations won and share of voice as leading indicators, connect them to AI-referred visits and leads wherever those are visible, and do not claim revenue we cannot trace.
What is share of voice in AI answers?
Your slice of the answers in your category: of all the citations that go to you and the named competitors across your prompt set, the percentage that goes to you. It tells you who leads your category inside AI engines, and by how much.
Can I see the raw data?
Yes. Every report ships with the prompt list, the per-engine results and the scoring. You can re-run any prompt yourself and check us. That is the point.
Want this run on your category?
A 20-minute intro call, no deck, no pressure.