Links and authority in GEO is the work of appearing in the places an AI engine looks when it decides whether to name you: the entity records it verifies against, the directories and review platforms it trusts, the roundups and comparison pages it cites, and the communities it reads. It overlaps with link building and then departs from it. The target list is not chosen by domain authority; it is read off the sources the engines cited in your own category answers, logged during the audit and re-logged at every review.
It is the slowest stage in the loop, because it moves at the speed other people publish, and it is the stage a monitoring tool or an in-house team most often cannot do at all. That is also why it is where most of the remaining points sit once a site is readable: above a health score of 40, the points left are mostly held by third parties (what moving one band takes).
why the entity comes before the links
A citation on a brand a model cannot verify does not hold; the name appears in an answer, then drops out. Across the 100 audits in our published dataset (April to July 2026), close to 100% of sites had no entity anchor, and not one had a Wikidata record. So the first half of this stage is records rather than links: a Wikidata item, one fact set across Crunchbase, LinkedIn and the directories, verified reviews on the platforms the engines read, and the corrections that resolve a name collision. In the INC4 case that meant Wikidata entries for the firm and both founders while Google's AI Overview was still resolving the name to an IEEE conference (the case, snapshot 27 July 2026).
what you get
| no. | deliverable | what is inside |
|---|---|---|
| 01 | the source map | Every source the engines cited in your category answers, from the audit panel, ranked by how often and on which engines. The list we work from, and the list you can check against the logs. |
| 02 | entity records | Wikidata, and for mature brands a Wikipedia route; Crunchbase, LinkedIn and the directories reconciled to one fact set: name, founding year, description, location. |
| 03 | review surfaces | Profiles claimed and cleaned on the platforms the engines read, and verified client reviews requested with the client's consent. Never written by us, never commissioned. |
| 04 | placements | Entries in the roundups, comparison pages, industry lists and communities on the source map, earned with the material from the content stage: a dataset, a case, a comparison someone else wants to cite. |
| 05 | backlink hygiene | The links you already have, pointed at pages that still exist. In the INC4 audit 177 of 549 backlinks pointed at dead pages; 41 redirect rules closed them (baseline 14 May 2026). |
| 06 | the citation log | Per review: which sources the engines cited about you, which were yours, which were independent, and which on the source map you have entered since the baseline. |
what we will not do
This is the stage where the shortcuts live, and every one of them has shown up in our audits as the reason an engine warned a buyer off a brand.
- No bought mentions sold as authority. In one audit the only readable corpus about a brand was its own paid promotion; the engines flagged the sources as promotional and issued a high-risk verdict. Volume without independent confirmation makes answers worse.
- No commissioned reviews. Reviews are requested from real clients, with their consent, on platforms that verify them. Our own first reviews on Clutch and Trustpilot were left by the client of a named case, August 2026, and say so.
- No suppressing accurate complaints. If a complaint thread is true, the fix is to publish the verifiable version of your side in machine-readable form, not to bury the thread. Engines read both.
- No private blog networks, no link schemes. They are the fastest way to have a domain discounted, and a discounted domain is invisible to this whole method.
how it is measured
On the same re-run panel as everything else, with one extra column: the source. For every category answer we log which pages the engine leaned on, and the delta that matters for this stage is the share of those that are independent of you and the number of source-map entries you have entered since the baseline. When the citation comes from a directory profile we cleaned, that counts; when it comes from a roundup you were not in three months ago, that counts more. Entity resolution is measured separately, by asking each engine to describe the company and checking every fact it states.
We also publish the honest version of this measurement for ourselves. Our own re-audit on 22 August 2026 scored 72 out of 100, up from 59 on 1 August, and the on-site layer explains the gap; the part that has not moved is this stage. Editorial links were still zero and one trust scanner still held a cached verdict from the domain's previous life (the about page). That is what the slow half of GEO looks like from the inside, and it is why the timeline for this stage is stated in months.
what it costs
Stage five is delivered inside the full-cycle retainer, from $3,000 per month (September 2026), with the final figure set after the audit. It is not sold as a link package, and there is no per-link price, because the value of a placement is whether an engine cites it, which nobody can price in advance. Placements are earned with material, not bought with budget, so the cost driver here is the number of markets, each of which has its own source map, not a volume of links. The drivers are on the pricing page.
where it sits in the loop
After technical GEO, because a source that links to a page the crawlers cannot read is wasted, and in parallel with content and SXO, which produces the material sources cite. The target list comes from the strategy. The whole loop is on the services page.