GEO-093·Issue 08·Technical Foundation·93 / 132
Structured Off-Site Profiles: Crunchbase, LinkedIn, G2
Crunchbase, LinkedIn, and review-site profiles are the brand fields AI checks first: twelve fields to align, five ordered steps, and one owner.
Issue focusOff-Site Authority: Digital PR, Communities, and the Knowledge Graph
Consider a composite case. A twelve-person espresso brand sells direct and is now taking café accounts around Austin. A café owner asks an assistant which machine brands can be leased for 24 months with on-site service there. Three come back; the first is introduced by its headquarters, founding year, and category. The brand’s entry is one clause: “also offers commercial espresso machines.”
The team pulls its off-site record. The site says founded 2016, Austin. LinkedIn still carries Amsterdam, where the founders incorporated. Crunchbase holds the pre-rebrand name and the industry “consumer electronics.” Two cities, two years, two names. The model used none of them.
Structured off-site profiles are the registration card AI reads first: the fields have to be complete and identical everywhere, because one platform with a different founding year gets the whole brand demoted to “sells this category too.”
Profiles are read as fields, not as introductions
Profile sites are pulled first because their content sits in fixed slots, not because the domains carry weight. An About page can say “we are obsessed with extraction at pressure”; no field receives it. LinkedIn has a founding-year field holding one value; Crunchbase holds one city.
Those slots also make the cross-check work: identical values on three sites read as fact, different values as a problem.
Entity work asks whether the brand exists; profile work asks whether its attributes were registered. Inside the off-site trust network, profiles are the base layer, not the volume layer.
A conflict is dropped, not averaged
Two things happen when fields disagree: the model demotes the attribute to “unclear,” or keeps the version most like a registry entry. It will not split Amsterdam and Austin, nor prefer the earlier founding year.
The worst position is not an empty profile. With none, a model can still quote the site; with conflicting profiles, the site becomes one more unverified version and the answer retreats to the safest true sentence: this brand also sells this category. Renames and moves produce this: the site changes, LinkedIn does not, Crunchbase keeps the old name.
A counterexample: six platforms registered, three headquarters cities, two brand names, and another quarter spent reconciling. Another: ad copy in a profile, “the leading end-to-end partner for modern coffee programs” — defensible, uncitable.
Three profile types, three profiles first
Pick platforms by the question AI uses them to answer, not by domain authority. One per type; a fourth waits for three to agree.
| Profile type | Examples | Question it answers | If it is wrong |
|---|---|---|---|
| Company record | Crunchbase, LinkedIn | Which country, how long, how large, what industry | HQ or year conflict; field unclear |
| Reviews | G2 and similar, trade outlets | Who uses it, what it is compared with, complaints | Wrong category; wrong candidate set |
| Directories | Trade shows, associations, regional B2B | Local service, local supply | Absent for “near me” |
G2 fits software and B2B services. Shipping physical goods, make that row a trade review outlet or category roundup, filling “who uses this and what breaks.” Ordering the three is a budgeting call, covered by prioritizing off-site investment; profiles are the cheapest lane, editable today with no development slot.
Five steps to one version of the truth
Copy the current state first. Twelve fields, a row each, a column per platform: name, founding year, headquarters, industry category, product category, size band, website, one-line description, markets served, main customer type, contact route, certifications. Empty means missing; disagreement is red.
| Step | Action | Owner | Passes when |
|---|---|---|---|
| 1 Copy | Twelve fields, every profile, one sheet | Content ops | Every cell valued or flagged |
| 2 Fix master data | Website wins; fix the values | Brand, product | One field sheet others can cite |
| 3 Write the description | 60–80 words, sourced like the answer unit | Content | Cover name; industry and buyer clear |
| 4 Resolve conflicts | Edit each platform; log what cannot change | Ops | Zero red cells, or a dated exception |
| 5 Maintain on trigger | Rename, HQ move, new category, certification | Brand lead | Four sources agree within 30 days |
Step 3 is where teams spoil it. A description is not a slogan; it answers who it serves, what it provides, for which market. Reuse it on directories and dealer kits with one checkable anchor: the site address or a certification number.
Example: one brand, four versions of itself
Example — company, values, and figures are invented; field names follow public profile pages.
| Field | Site | Crunchbase | Review site | |
|---|---|---|---|---|
| Headquarters | Austin, US | Amsterdam, NL | Amsterdam, NL | blank |
| Founded | 2016 | 2015 | 2013 | blank |
| Industry | Commercial espresso equipment | Machinery | Manufacturing | Food service software |
After alignment: Austin, Texas, US; founded 2016; commercial espresso equipment, lease and sale, B2B; one description everywhere — “Leases and services commercial espresso machines for cafés, offices, and restaurant groups in the US, with installation and service.”
Nothing was added; every cell set to one value, so a cell lifted alone stops arguing with another. The category cell matters most: a wrong category seats the brand in a candidate set no buyer asks about. Results show in citation share on brand-fact prompts: answers start carrying the founding year and city, not only the name.
Where this breaks
Marketplace-only sellers. If checkout happens on Amazon, platform AI reads listings, reviews, and fulfillment. Profiles trail that work.
Pure DTC consumer brands. No B2B procurement or dealers: the Crunchbase row answers a question your buyer never asks. Put those hours into reviews and UGC.
Review sites built on verified users. Bought reviews break the rules and leave a trail; when reviews stop matching the product, buyers notice. Leave the slot empty.
Two misuses travel with this. Treating a profile as the website: it says who you are, not whether a machine fits under a 17-inch cabinet. And treating it as a one-off: a year later there are three versions again.
Profiles sit at the cheap end of the technical foundations hub: no development queue, no budget approval, one named owner for the fields.
This month, do one thing. Open the About page, the LinkedIn page, and the Crunchbase record; copy founding year, headquarters city, and industry category into three columns. Fix the disagreeing field first, then decide about a fourth platform.