Unstructured citations and why niche local blogs matter

That difference is the whole article. A listing asserts that a business named X exists at an address with a phone number. A neighborhood blog asserts that X is the one with the patio, that the owner used to cook at the place on Kapahulu, that they open early on Sundays for the running club.

One of those is a record. The other is a description. Systems assembling answers about local businesses need descriptions, because the questions people ask are descriptive.

What actually counts as an unstructured citation?

Any mention of your business in running text. A local news story. A chamber of commerce roundup. A hobbyist forum thread. A wedding blog listing venues. A school newsletter thanking sponsors. A subreddit answering "where do I get this fixed in Alameda."

No NAP block, no consistent formatting, frequently no link at all. Just your name in a sentence written by someone who is not you.

The value is in the sentence, not the link. A link is a bonus. What the mention contributes is a specific claim about your business, published somewhere other than your own site, that a retrieval system can use when a question requires that exact claim.

The best available read on whether that actually happens comes from Whitespark's testing of AI Mode local answers in May 2026, which found reviews and unstructured citations, meaning blogs, news and community pages, to be the heaviest inputs to those answers. That is practitioner observation rather than a controlled test, so hold it loosely and note that it comes from a company selling local SEO tools. It is still the most specific published statement anyone has made about this mechanism.

There is a quieter signal pointing the same direction. In Whitespark's 2026 Local Search Ranking Factors survey, where 47 local search experts scored 187 factors, structured citations lost about half a percentage point of weight, continuing a multiyear decline. The formatted directory listing is fading as the unformatted mention rises. Those are two different things that share a word, which is most of why the category is confusing.

This is also a different thing from structured markup, and the two get confused constantly. Schema tells a machine how to parse your own page, and Google is explicit that it is not required for AI features. A citation in a local blog is somebody else vouching that you exist and what you are like. Only one of those is hard to fake.

Why would a small blog beat a big directory?

Because information value scales with rarity, not with traffic.

A national directory has a page about your business that is structurally identical to its page about every other business in your category. There is nothing in it that a retrieval system could not get from a hundred other sources. It adds volume, not information.

A neighborhood food blog comparing three ramen shops by name, with opinions about each, is producing text that exists nowhere else on the internet. When someone asks an assistant which of those three is best for a quick lunch, that post is the source that can answer.

The system is not measuring authority the way a link-building deck would. It is looking for text that addresses the question, and a niche post about your specific street addresses questions that a directory listing cannot.

There is a compounding effect with your reviews here too, since both are third-party text describing you in specific terms, and a review cadence that has gone quiet removes half of that supply.

Where does this mechanism break down?

At the point where it works too well, which is the part vendors selling mentions do not mention.

Ahrefs built a fictional luxury brand, verified the name returned zero Google results, generated 56 adversarial questions using a different model from the ones being tested, and put eight systems through it. With only the brand's own site published, the strongest models answered 53 or 54 of the 56 correctly. Perplexity failed roughly 40% of them, at one point insisting the brand made smartphones.

Then the second phase. Ahrefs seeded three mutually contradictory fake sources, a blog post, a Reddit thread and a Medium investigation, while leaving an official page on the brand's own site explicitly denying the claims. Perplexity and Grok repeated the fabrications as fact. Gemini and Google AI Mode flipped from skeptical to convinced. The brand's own denial lost to the third-party consensus.

Read that as a warning label on the whole mechanism. These systems treat repetition and narrative detail across several independent-looking sources as evidence, and they will override a business's own first-party statement to do it. That is what makes an earned local mention valuable. It is also what makes the manufactured version dangerous, and it is why a competitor's smear or a stale fact about you can outlive your correction.

One combination that is not available. Perplexity's roughly 40% error rate came from a fictional-brand test on one engine, and the finding that unstructured citations are heavy inputs came from a different company's local testing on a different surface. Multiplying them into a probability that your AI answers are wrong is not a calculation anyone can do. Neither study publishes the joint distribution and the samples do not overlap. Report each next to what it measured.

Ahrefs sells SEO software, so grade the experiment as vendor research. The design is disclosed and it is a genuine controlled experiment, which is rare enough in this category to be worth saying.

What is a brand mention package actually worth?

Less than the invoice, and the downside is not priced in at all.

The quote. A vendor offers "brand mention building for AI visibility" at $2,000 a month for fifteen placements. That is $24,000 a year and about $133 per placement (derived).

What Google says about the product. It names "seeking inauthentic 'mentions'" directly as ineffective and spam-policed, alongside special files, artificial content chunking and content rewritten for AI systems. So the seller is offering the imitation of a mechanism, described using the mechanism's own evidence.

What $133 actually buys. The seeded sources in the Ahrefs experiment worked because they were detailed, narrative-rich and mutually corroborating across three formats. That is expensive to produce. At $133 a placement you are buying a templated paragraph on a site nobody reads, which is the version that does not clear the bar the experiment set.

Now price the downside, which is the part nobody quotes. A burned account and a banned domain in your own city's subreddit is a permanent loss in the exact community you sell into. In regulated categories, a fabricated testimonial is a violation of the federal rule on consumer reviews, which carries civil penalties for knowing violations. Check the current-year penalty figure before putting a number on it, but note the structure: the FTC can seek civil penalties for a first-time violation, which it generally could not before October 2024.

The alternative at the same budget. One local sponsorship at $500, a standing rule that you answer a reporter within the hour, one community event, and two hours a month writing down an opinion with your name on it. Call the cash cost $2,000 for the year (derived), leaving $22,000 derived that never leaves the business. The placements are fewer. They are also real, they are in publications your customers read, and no platform policy names them.

The honest limit. Earned mentions are slower and you cannot forecast the count, which is exactly why the purchased version sells. If you need fifteen mentions this month, you do not have a marketing problem, you have a deadline someone else set.

How do you earn one without buying it?

Slowly and locally, which is the part nobody wants to hear.

Sponsor something small and local where the sponsor list gets published. Give a real quote to a reporter covering your category, quickly, because reporters remember who answers on deadline. Host or supply an event that a community calendar lists. Answer a question in a local forum as yourself, disclosing what you do, without a pitch. Hire someone notable enough locally that it gets written up. Have a genuine opinion about something in your industry and put your name on it.

Each of those produces one sentence about you somewhere real. That is the unit of work. It is slow, which is consistent with how long local SEO takes before anything moves, and it does not compress.

Does the mention need to describe you consistently?

Yes, and this is where most of the value gets lost.

If one blog calls you Kalani's Auto, another calls you Kalani Auto Repair LLC, and a third lists a phone number you stopped using in 2023, you have produced three weak signals instead of one strong one. The system has to decide whether these are the same business. The Perplexity result above is what that decision looks like when it fails, and that failure happened on a name with no conflicting information anywhere on the internet.

Before you go earning mentions, make sure the facts a writer will copy are correct in the places they will copy them from. Entity consistency across the web is the unglamorous prerequisite that makes every later mention count.

And set expectations on the payoff. Getting cited in an AI answer is not the same as getting a visit, since AI Overviews changed what a top ranking is worth in clicks. Mentions are worth earning anyway. Just measure the right thing.

What to do this week

1. Search your business name in quotes and read every result that is not a directory. That is your current citation set. It is usually shorter than owners expect. 2. Read those results for factual errors about you. Old address, old hours, a service you dropped. Corrections are cheaper than mentions and worth more. 3. List five local publications, forums or newsletters that already cover your category or your neighborhood. 4. Pick one and find a legitimate reason to be in it this quarter. An event, a hire, a seasonal angle, an opinion. 5. Write two sentences describing your business in the words you want repeated, and use those exact sentences everywhere. Writers copy. 6. Confirm your name, address and phone are identical on your site and your Business Profile before anyone else copies them.

Want a list of publications worth being in?

Send your city and category to eric@seod.com and I will send back five local publications, forums or newsletters in your area that already cover businesses like yours, with a note on how each one tends to get its stories. If you have a mention-building quote in hand, send that too and I will tell you what the placements actually are. No charge and no pitch.

I do this because the research is the hard part and the outreach is not, and most owners stall at the research.

Otherwise, there is more on AI search and AEO here.