Consensus Is the Hardest AI Visibility Score to Move
Brands that carry both mentions and citations in AI answers are 40 percent likelier to resurface across consecutive queries than brands with citations alone, according to the 2026 AI Visibility Index. Repetition across queries is what turns a lucky mention into reliable visibility.
Third Hemisphere scores AI visibility with CRISP, on five factors: Consensus, Reputation, Influence, Source, and Presence. Consensus rests on sources the company does not control, which is why it takes the longest to move and produces the most durable result.
What is Consensus in AI visibility?
In CRISP, Consensus asks whether the engines agree about a company: when one engine names it on a question, whether the others answering that question name it too. That agreement is built on the open web. It depends on whether the company's own site, media coverage, industry directories, analyst notes, partner pages, and customer references all define the organisation, its category, and its products in compatible terms.
An AI engine synthesising an answer is reconciling multiple retrieved sources. Where those sources agree, the model can state a claim with confidence and attribute it. Where they conflict, the safest output is a vaguer sentence, a competitor whose description is cleaner, or no mention of the company at all.
The other four factors ask different questions. Presence asks whether the engines reach the company when a buyer describes the need without naming it. Reputation asks whether the engines put the company forward. Influence asks whether the company's account of itself is what gets repeated, and whether anyone else carries it. Source asks whether what the company publishes stands up as the record of what is true about it. Consensus depends on a harder question: do independent parties, none of whom the company controls, tell the same story.
Why is Consensus hard to move?
Four reasons.
The sources belong to other people. A company can rewrite its own pages this afternoon. It cannot rewrite a 2023 trade article, a directory entry created by a former employee, or a partner's product page.
Old descriptions persist. Companies change categories, drop product lines, and reposition. The web retains every previous version, and retrieval systems do not automatically prefer the newest.
Internal inconsistency propagates. Where a company has described itself three ways across five years, third parties have copied all three. Fixing the source does not fix the copies.
The fix requires earned outcomes. Building agreement means securing new coverage, new listings, and new references that use the current description. That is media relations work, and media relations runs on other people's timelines.
How do you diagnose a Consensus problem?
Three checks surface most of it.
Ask five engines to describe the company in one sentence. Compare the five answers. Divergence between engines is a Consensus signal, because each engine has retrieved a different subset of the same conflicting sources.
Search the company name and read the first 20 results as a set. Count how many distinct category descriptions appear. Two is workable. Four is a problem.
Check the oldest well-ranked source. A high-authority page from four years ago describing a product the company no longer sells will outweigh three recent blog posts.
Where does the conflict usually come from?
Five sources produce most of the divergence, and none of them are malicious.
Repositioning. A company that moved from one category to an adjacent one leaves a complete, well-linked record of the old position behind it. Retrieval systems find both.
Funding announcements. Coverage of a raise is written quickly from a release, and the description in that release fixes how the company is characterised for years afterwards. A sloppy sentence in a 2022 announcement is still being repeated.
Directory entries. Startup databases, association registers, and procurement panels contain descriptions submitted years ago by someone who has left. They rank well and nobody reviews them.
Partner and reseller pages. Partners write their own descriptions of what a company does, usually optimised for their own positioning, not the vendor's.
Executive bios. A founder's conference biography, podcast description, and LinkedIn summary frequently describe the company three different ways, and all three are retrieved.
What actually moves Consensus?
Five actions, in rough order of effect.
Action/Why it works
Fix the canonical description: Agree one sentence describing the company and use it verbatim everywhere the company controls, including releases, bios, directory entries, and partner submissions.
Earn new coverage using it: Journalists paraphrase from the material supplied. A consistent description in the boilerplate and the pitch propagates into independent sources at no extra cost.
Correct high-authority stale sources: Directories, association listings, and partner pages usually accept an update request. Each correction removes a competing version.
Publish customer and partner evidence: A customer describing the company in their own words, using the current category language, is the strongest independent agreement available.
Keep spokesperson bios identical: Conference bios, podcast notes, and article bylines are retrieved frequently and drift constantly.
What does the canonical description need to contain?
One sentence, used without variation, doing five jobs at once.
The legal name in the form the company wants used everywhere.
The category, in the words buyers and journalists already use, with invented terms left out.
What the company actually does, stated concretely enough to be checkable.
Who it serves, narrow enough to be meaningful.
Where it operates, since geography is one of the details engines most often get wrong.
The temptation is to write something distinctive. Distinctiveness belongs in the sentence after it. The canonical sentence has one job, which is to be repeated accurately by people who are not paying close attention.
How long does it take?
Changes to material the company controls can register within a quarter. New credible third-party sources accumulate over two to three quarters. Consensus typically takes three to four quarters to show a clear change, and longer where the existing conflict is severe.
That timeline argues for starting with the canonical description before any campaign. Every piece of communications output published before the description is settled adds another version to the pile the company will later have to reconcile.
Who owns Consensus inside a company?
Nobody, in most organisations, which is the reason the score stays low.
The web team owns the site. The communications team owns coverage. Partner marketing owns partner pages. Talent acquisition owns the employer profiles. Each function writes its own description of the company, all of them defensible, none of them identical, and no single person reads the set.
The fix is administrative. One person holds the canonical sentence, every function uses it, and any request to change it goes through that person. Companies that install this find their Consensus score moves within two quarters, without a single additional placement.
What does this mean for the media programme?
It reframes what a placement is worth. Coverage has always been valued on reach, outlet quality, and message pull-through. A Consensus lens adds a fourth measure: whether the piece describes the company in the terms the company wants repeated.
A large feature that describes the business inaccurately can lower Consensus while raising Presence. A modest trade article that uses the exact category language raises both. Neither outcome is visible in a coverage report that counts placements and audience figures.
The practical change is small and awkward. Someone has to read each placement against the canonical description, log where it diverges, and feed that back into the next round of pitching. Most agencies do not do this. It is also the habit with the largest effect for a communications team that cares about AI visibility.
What does a low Consensus score cost?
It costs inclusion. An engine that cannot settle what a company does will produce an answer naming the competitors it can describe cleanly, and the company never appears in the shortlist that forms before a buyer visits any website.
It also costs accuracy where the company does appear. Conflicting sources produce hedged, generic, or outdated descriptions, so a company that repositioned two years ago finds itself introduced to buyers by its old category.
The compounding effect is the expensive part. Presence improves faster when Consensus is high, because consistent sources reinforce each other. A company fixing Consensus is raising the ceiling on Presence at the same time.
Where does this sit in AI Authority & Visibility?
Third Hemisphere runs this work through its AI Authority & Visibility practice. It takes a company's existing marketing strategy as the input and assesses each area of it for AI visibility. It then runs a CRISP audit, asking 92 questions about the company by name across ChatGPT, Claude, Copilot, Gemini, and Perplexity, and scores the results on Consensus, Reputation, Influence, Source, and Presence. The practice sits as a strategy layer across a company's in-house team and the agencies it already retains, which is necessary for Consensus specifically, because the fix requires the web team, the media programme, and the partner marketing function to publish the same sentence.
The agency's Generative Engine Optimisation practice covers the citation side of the same problem.
The takeaway
Consensus measures whether AI engines agree about a company, and that agreement rests on independent sources describing it the same way. It decides whether an AI engine repeats a company across related questions or mentions it once and moves on. Fixing it starts with one agreed sentence, propagates through earned coverage and corrected listings, and takes three to four quarters to register. Companies that want their five CRISP factors measured can start with AI Authority & Visibility or get in touch.