Sentiment Delta and What AI Says About You
I measure the gap between the positioning I intend and the description AI engines actually return, and I spend to close that gap before I spend to be mentioned more often. The gap is sentiment delta. On the queries where a buyer is deciding, the engine's sentence about us is the first thing they read. Showing up inside the wrong sentence is a revenue problem. I test that on our own checkout.
The definition and the four-type method are in What Is Sentiment Delta? | MR Research. The queries I score, and the sources I hold responsible, are below.
Four gaps, kept separate
Sentiment delta is the distance between intended positioning and the description an engine returns. The research page separates four types, and I log them as four rows.
Directional: the frame points the wrong way. The company says enterprise-grade. ChatGPT says it is a fit for small teams.
Categorical: the market is wrong. The company sells measurement for communications. Perplexity files it under social analytics.
Omission: the brand is absent on a query it should be able to answer. Founders notice this one because it is binary.
Attribute: the emphasis is wrong. The differentiator is public pricing. Gemini leads with a Salesforce integration.
Zero gap is rare. Muck Rack's May 2026 edition of "What Is AI Reading?" put earned media, sources brands neither own nor pay for, at 84% of the more than 25 million links it analyzed (Muck Rack). That counts all citations, and it is a vendor's own category. I treat the About page as our draft and the earned record as the draft the engine leans on. The names for these gaps sit in the glossary. The sequence of authority, entity, and measurement around them is on Machine Relations. I coined Machine Relations in 2024 so this work would have a name a budget could attach to. Sentiment delta is one of the three metrics I report with it, beside share of citation and whether the entity resolves. Share of citation says how often we appear. Resolution says whether the engine has the right company. Sentiment delta says what it says when it does.
What I can measure, and what I will not quote
I do not have a verified conversion multiple for a false or negative answer. Vendor write-ups give one, built on joining AI sessions to checkouts, and their method is not retrievable, so I am not quoting them. A buyer asking "is X worth it" or "X versus Y" is close to a decision, and a false sentence in that answer is a false sentence addressed to that buyer. I measure the leak on our own sessions: AI-referred evaluative visits, split by what the answer said, against checkouts or demo requests. Until that exists, the ordering below is judgment, and the board note says so.
Another post on our domain is the weak fix
Ahrefs, on 75,000 brands, found branded web mentions correlating at 0.664 with appearance in AI Overviews, against 0.218 for backlinks, Spearman. Mentions were the strongest of the factors in that study and backlinks among the weakest. Ahrefs' follow-up on the same pool of brands, across ChatGPT, AI Mode, and AI Overviews, found branded web mentions still correlating at 0.66 to 0.71 and the number of pages on a site at about 0.19 (Ahrefs). Its strongest factor there was YouTube mentions, at about 0.74. Ahrefs says correlation is not causation, and the coefficients are not a ratio. Between another backlink campaign and another post, I fund mentions.
FrictionAI separates recognition from recommendation. In its sample, New Balance had the strongest Knowledge Graph entry and was recommended for athleisure 3.4% of the time, while Lululemon, with one seventy-ninth of that Knowledge Graph score, was recommended 92.5% of the time. That is one vendor's study in one category. Knowledge Graph score is a poor proxy for being the recommended vendor. Recognition is the graph. Recommendation is what other people wrote. A wide sentiment gap is a signal that we are far from the sources engines prefer in our own category. Our website, alone, leaves the gap open.
The audit before the quarter's plan
I pick five queries a buyer asks before they know our name. Best in the category for the use case. Whether the product is worth it. Us against alternatives. The category's actual buying question. I skip the brand-name query for this audit, because omission and distortion on the way in are the expensive rows.
I run each query on ChatGPT, Perplexity, Claude, and Google AI Overviews, more than once and on more than one day. A Swiss-German repeated-measurement study found that cited sources change from day to day (arXiv), and one engine on one afternoon is a sample of one.
I score the language. Plus 2, strongly positive, recommended as the choice. Plus 1, positive with a minor caveat. Zero, balanced. Minus 1, a real caveat in the lead. Minus 2, strongly negative or false. Then I compare the score with the positioning we claim. "Most transparent pricing" against "pricing not publicly available" is a directional gap. I write the sentence down.
For every zero or negative, I open the cited URLs. If the engine is citing a page that frames us wrongly, that page is the remediation target. If the engine cites a page that never names us, that page is a reason to get named inside a source the engine already trusts.
The pass takes about half an hour. I do it before I approve another month of undifferentiated content.
What I fund to close a row
The spend order puts third-party mentions first, on the Ahrefs correlations, news and trade publications next, and owned content third. The words a third party uses, the attributes they name, and whether they recommend, are inputs to the answer. That is my judgment from how engines quote what they retrieve. A generic mention leaves the gap open. A mention that states the attribute we are mis-scored on is the one I pitch.
My cadence: weekly for four weeks after a placement, scored against the specific query and the specific gap it was bought to close, then monthly in a steady state, and daily during a live reputation event. I do not have a verified timeline for earned media to move an answer, so I set a four-week check per placement and a quarterly review, and I report what moved.
Owned content still has a job. It is the page the third-party claim should agree with. It is a weak tool for overwriting a third party the engine already prefers.
Which gap gets the money first
A false or strongly negative sentence comes first. It is a wrong statement in front of a buyer who is deciding, and I correct it before I know the size of the leak. An omission on a query nobody was ready to buy from waits behind it.
Directional and attribute gaps are the next invoices, and they are earned-media invoices. I pitch the attribute we are mis-scored on, in a publication the engine already cites, and I score the same query weekly for four weeks.
Categorical error is an entity problem wearing a sentiment score. If Perplexity has us in the wrong market, more adjectives in the right market will leave the file where it is. The third-party sentence has to name the category, and our own pages have to say the same category. FrictionAI's split is why a Knowledge Graph gain is a substitute I decline.
Omission is what I measure and refuse to panic-buy. Silence converts nothing from that answer. A present, false answer can teach the buyer the wrong lesson. Accurate description is the goal. Empty presence is a purchase I skip.
The decision
I score the five buyer queries across the four engines, log each gap as directional, categorical, omission, or attribute, and fund earned coverage that names the attribute we are getting wrong. Any program whose only success metric is mention rate waits until the evaluative-query scores are at least neutral.
The scope is evaluative queries, the four engines above, and the cited URLs behind any score of zero or below. The Ahrefs results are why the spend goes to third parties, as correlation. The review horizon is four weeks per placement and a quarter overall.
An evaluative-query sample on our sessions, in which strongly negative answers convert at or above the positive baseline, overrides the ordering and keeps the audit, because a false description is still a false description. The owned-content rule changes if a controlled change on our domain, with no new third-party coverage, moves the score on more than one engine across repeated runs. A Knowledge Graph increase that left recommendation rate near FrictionAI's 3.4% example would leave the gap open.
FAQ
What goes on the monthly scorecard?
The four gap types, the score on each engine for the five queries, and the URL behind any negative. Share of citation can rise while the description stays wrong. A presence chart does not retire this one. The research page is the method. The scorecard is the log of our queries.
Does sentiment move revenue?
I do not have a verified number, and I will not borrow one. I join AI-referred evaluative sessions to checkouts or demo requests on our own site, split by what the answer said, and read the difference. A sentence that is simply false gets fixed while that measurement is being built.
How long before a placement is judged?
Weekly for four weeks after the placement, against the specific query and the specific gap it was bought to close. A single engine moving on a single run is a lead.
What if we are absent rather than described badly?
That is omission, and it is a different row from a directional gap. Absence converts nothing from that answer. A present, false description can be worse than silence on a query where the buyer is ready to purchase. Accurate description is still the goal. Empty mentions are a purchase I skip.
About the author
Jaxon Parrott founded AuthorityTech in 2018 and coined Machine Relations in 2024. He writes about earned authority, AI discovery and founder judgment.
Continue reading
The Category Ownership Trap
Owning a category name leaves the buyer's question open. I coined Machine Relations in 2024, and the commercial question is still a separate job.
What to Do When AI Search Ends the Click
I manage growth to whether an AI answer cites the company, because a search that ends without a click never reaches the site.
How to Hire a GEO Agency
I hire a GEO agency to make pages extractable, and buy earned coverage when non-owned, non-paid sources carry 82 to 89 percent of AI-cited links.