For 20 years the conversation about anchor text was about PageRank. Exact match, partial match, branded, naked URL. The goal was to push ranking signals without tripping a spam penalty.

That conversation still exists for classical Google. It is no longer the most important conversation about anchors. In AI search anchors do a different job — they help the retriever resolve which entity your domain represents.

We learned this the awkward way when Answerly Agency launched and AI Overviews kept describing answerly.ai (a different product) when someone searched our brand. Same root name, very different companies. The fix was not more content. It was anchor text.

The entity-resolution problem

Open ChatGPT and ask “tell me about Answerly.” Until very recently — and still on some prompts — you would get a description of an AI customer-support product. That product is real. It is also not us.

The LLM has to resolve which entity the prompt is about. It does this by looking at how other documents on the web refer to “Answerly.” If most of the references point at the customer-support product, the customer-support product wins the entity slot. Our domain (answerly.agency) gets ignored or merged into the wrong description.

This is not a content problem on our site. We could write a hundred more articles and not fix it. It is an off-site signal problem.

The LLM does not read backlinks the way Google does. It does not assign PageRank weight per anchor. But it does read the anchor text of inbound links as a signal of how the world describes the linked entity.

If 40 inbound links to answerly.agency use the anchor “Answerly” and 80 inbound links to answerly.ai also use “Answerly,” the LLM has 120 references to “Answerly” with no way to tell which domain represents which entity. It guesses. It usually guesses wrong, because answerly.ai has been around longer and has more references overall.

If those 40 inbound links to us instead use “Answerly Agency” — and the 80 inbound to them keep using “Answerly” or “Answerly.ai” — the LLM has 40 references to one entity (“Answerly Agency”) and 80 to another (“Answerly” / “Answerly.ai”). The disambiguation works. Our domain gets associated with the correct entity. AI answers stop describing the wrong product.

This is the entire move. Stop letting LLMs read your domain through the same string as a competing brand.

The “brand + category” rule

The pattern that wins is anchor text that combines the brand name with the category. Some examples from our work and partner sites.

  • “Answerly Agency” — clean
  • “Answerly Agency (AEO consultancy)” — better, adds the category in parentheses
  • “AEO services from Answerly Agency” — better still, full descriptive phrase
  • “the AEO arm of Chyzh Agency, Answerly Agency” — long but unambiguous

The pattern that loses.

  • “Answerly” — string collision with answerly.ai
  • “Click here” — useless for entity resolution
  • “AEO services” without the brand — passes ranking weight but does not resolve the entity

We rewrote 14 inbound links across chyzh.agency, partner sites and our own outreach drafts. Most went from “Answerly” to “Answerly Agency.” Within six weeks AI Overviews started correctly describing our agency on brand-search prompts. The two competing entities now resolve cleanly.

What about exact-match keyword anchors

Still useful — for classical Google ranking and for indicating topical relevance to the retriever. Keep around 15–25% of your inbound profile on keyword anchors when those anchors come from contextually relevant sources.

What changed is the priority order. In 2018 you would say “build the keyword anchors carefully, sprinkle brand for variety.” In 2026 it is “anchor the brand correctly first, build keyword anchors where it makes sense contextually.”

For a new brand competing with a same-root incumbent, the brand+category work is the first six months. The keyword work comes after the entity layer is solid.

Anchor text is one input. The retriever reads several others. Together they form what the model treats as the “entity record” for your domain.

Wikidata. A free record on Wikidata is the highest-leverage single move you can make. The retriever reads Wikidata to resolve named entities. If you have a Wikidata entry that says “Answerly Agency is an AEO/GEO consultancy in Kyiv, distinct from the unrelated Answerly product (Q12345)” — the model now has a structured way to keep the two separate. We submitted ours the week we launched.

LinkedIn Company page. A complete LinkedIn page with a clear tagline, description and named team members carries roughly the weight of three or four mid-quality backlinks for entity resolution purposes. The model reads it directly.

Crunchbase, Pitchbook, GitHub Organization. Each adds a slightly different angle. Crunchbase confirms the company’s founding and category. GitHub confirms which projects belong to which entity. The model reads all of them when entity resolution is in doubt.

Schema.org sameAs array. On the Organization schema of your own site, the sameAs array points to the canonical references — LinkedIn, Wikidata, Crunchbase, the parent agency page. We have written about this in detail in the schema stack for AI citation post.

The combination — clean anchor text + Wikidata + LinkedIn + sameAs — is what resolves the entity. Backlink anchors alone are necessary but not sufficient.

What to do in outreach today

Three concrete changes to outbound link requests.

Specify the anchor in the request. When emailing a partner about a link placement, do not leave the anchor up to the editor. Say “please anchor it as ‘Answerly Agency’ (not bare ‘Answerly’).” Most editors will comply because it is easier than picking on their own.

Audit your existing inbound links. Pull your top 50 inbound links — by referring domain authority, not by count. Look at the anchor on each one. Anywhere it is the bare brand string with an entity collision risk, ask politely for an edit. We had 11 of our top 14 changed by emailing the editors with the reason.

Train your team on the new pattern. Brand mentions in guest posts, podcast bios, conference programmes — everywhere a writer types your brand. Default to brand + category. Make it the house style.

For an outreach script you can copy, see reddit citations and backlinks — the same principles apply on platforms that pass entity signals even when they do not pass classical link weight.

A note on the cost

This work is unglamorous. Rewriting anchors is not the kind of move that makes a quarterly slide deck look exciting. The lift is also slow — six weeks before AI answers visibly shift.

But it is unusually permanent. Anchor text on contextual placements does not decay the way citations do. Wikidata entries persist. LinkedIn descriptions persist. Once the entity record is correct, it stays correct, and every new piece of content you publish gets attributed to the right entity from the start.

For a new brand the entity layer is the first six months of off-site work. Skip it and you spend the next two years competing with a name collision the LLM never bothered to resolve.