CitePack studies one buying decision at a time: what ChatGPT, Claude, Gemini and Perplexity answer when asked what to buy, the work those answers point to, and what changed when the same questions are asked again. This post walks through how one Decision Loop is built and what each control is for.
One loop runs in this order:
- Day 0: 60 answers, read into a Decision Map, a Work Plan and an Action Pack.
- You act: the customer does the work and marks each item done.
- Measure Again: the same questions, 60 more answers, a second Decision Map, Work Plan and Action Pack, and a Before/After Snapshot of the two maps.
That is 120 answers in one loop. The examples below use Crosstrow, an illustrative company from the CitePack site, so none of its figures are measured results: https://citepack.com/example.
One offering, one category, one market
A study is fixed to one offering, one buying category and one buyer market, which is a country. Change any of the three and it is a different study with its own questions.
The market reaches every engine two ways. Each question names it, and each engine is set to it: ChatGPT, Claude and Perplexity through their own location setting, Gemini through a location point in that country. The customer can turn this off before approving, and then each engine answers from its own default. The map records which way each engine got the market.
Five questions, sealed before anything runs
CitePack drafts five questions, one for each buying situation: category consideration, job suitability, mandatory eligibility, balancing priorities and change of solution. Each can carry up to two buyer specifics, such as a must-have or the work to be done, or none when it should stay broad.
The customer edits any wording and approves all five. From then on they are sealed: every engine gets the same words, and Measure Again reuses them, so the two maps compare like with like.
Four engines, three clean repeats
Each engine is called through its official API with web search on, pinned to the model its free app gives buyers. Every answer records the model that wrote it.
Each question goes to each engine three times, with no account history or memory: 5 questions × 4 engines × 3 repeats = 60. One answer can't tell you whether the next will match. Three show where an answer holds and where it shifts, and all three are kept.
This also sets the limit. A person's own app is shaped by their history and settings, so one screenshot describes one person. A map describes the pattern across 60 controlled answers in one market.
What the map reads
An answer can name a company without putting it forward as an option, and cite a page without recommending its product. The map keeps mentioned, offered and cited apart.
When an answer lists options in a clear order, the map records each company's place: first, second, third. When the order is unclear, it says so instead of guessing.
Pages an engine cited and pages its searches found are shown separately. What the customer confirms, what its own pages say and what an answer says are never passed off as one another. Prices, free plans and amounts an answer states are checked against the company's website.
The Work Plan and the Action Pack
The Work Plan follows what the engines did: the searches they ran, the companies they looked up by name and the pages they read. CitePack reads those pages too and shows where the company and the company named first sit on each one.
Every item names its question, the answers behind it, the evidence word for word, its kind, its effort, its owner and the check to run at Measure Again. An item with no answers behind it doesn't go in. There are nine kinds of work, from getting onto a list AI reads to correcting what AI gets wrong.
The Action Pack holds the material for each item: an email draft, a page draft, a rewrite, a correction note or a technical fix, whichever the item calls for. Facts about the company are quoted from its own site, and nothing about a rival is stated for it.
How the Work Plan works: https://citepack.com/work-plan. An illustrative Work Plan: https://citepack.com/example-work-plan.
Measure Again and the Before/After Snapshot
On Day 28, or a day the customer chooses, the same sealed questions go to the same engines in the same market. Each question now has 24 answers across the loop.
The snapshot marks each engine Same or Different, with how many of its fifteen answers changed, and notes any engine whose model changed in between. Each Work Plan item shows its check, answered on the new map, beside whether it was done. The new map gets its own Work Plan and Action Pack.
In the Crosstrow example, one question was named first in 2 of 12 answers on Day 0 and 7 of 12 on Day 28. A map with no change is a result too.
What the loop does not claim
No single score, no league table and no continuous monitoring. It doesn't guess why an engine answered as it did, and it doesn't promise a recommendation, a citation or that any change will move the answers. It reports what was measured on each day.
Price
- Founding seat, first ten companies, first come, first served: one loop for $99, or ₹4,999 if you pay from India.
- After the ten: one loop for $149 or ₹7,499.
- Each further loop for existing customers: $99 or ₹4,999.
- Packages: email founder@citepack.com.
The map is ready 30 minutes on average after the questions are approved, with a full refund if it isn't delivered within 24 hours.
To try the method on one question first, the free sample runs one buying question on two engines: https://citepack.com/free-sample. The full method: https://citepack.com/methodology. Common questions: https://citepack.com/faq.
Top comments (0)