Liu Yu · Tianwen SEO Research Notebook
Why Self-Promotional Content Backfires with AI: A Lesson from One Experiment
“We are experienced, and our quality is reliable” — that eight-character phrase is the one I see most often on export sites, and the least useful. It is not wrong; it is empty. So empty that readers skip it and AI cannot cite it.
A few judgments to take away first
- Ahrefs ran an AI SEO experiment on self-promotional content, and the conclusion was in the title: self-promotion works — until it backfires.
- The mechanism is no mystery. Self-promotion supplies no new information; readers cannot verify it, and AI cannot take any usable fact from it.
- Ahrefs’ 2026-07-27 Reddit study confirms this from the other side: what search systems value is specific experience and real discussion.
- In our own T1 panel, the number of times we were named among providers went from 0 to 2, and both appeared under questions like “how to choose a service provider” — not because we praised ourselves.
- Replace adjectives with numbers, conditions, and failure records, and a page turns from self-promotion into evidence.
What have the authority sites measured recently?
Ahrefs published “Self-Promotional Content Works—Until It Backfires,” an AI SEO experiment they ran. The title itself sums up the conclusion: talking about yourself in moderation is useful; talking about yourself throughout the page crashes. They did not say “you cannot mention yourself”; they drew a boundary.
Where that boundary lies can in fact be reasoned out with common sense. A reader clicks into a page to solve their own problem. If your first screen is entirely about how impressive you are, they will decide “this is an ad” and leave. AI judges more directly: it has to give the user one usable answer, and if your page has only adjectives, there is no sentence to extract, so it has to look elsewhere.
The same firm’s 2026-07-27 Reddit study gives a reference point: Reddit appears in 83.9% of Google’s discussion and forum results. Reddit posts are usually not self-promotional; what they offer is specific experience. Search systems favor this kind of content not because they are anti-business, but because it answers questions better.
What are the three typical kinds of self-promotion on export sites?
| The self-promotional phrasing | What the reader gets | A verifiable replacement |
|---|---|---|
| “Experienced and industry-leading” | No information | “Since 2016 we have made only this type of part, with more than 400 containers shipped in total” |
| “Reliable quality, satisfied customers” | Cannot be verified | “Batch defect rates are sampled against standard X; two returns in the past two years” |
| “We have served many well-known clients” | Equivalent to saying nothing | “We have done OEM work for three European retail chains; one of them specified…” |
The third is the most troublesome. Many owners feel that naming clients involves confidentiality, so they write only “many well-known clients.” To a reader that is not caution, it is emptiness. In fact you do not have to name the client; you can write the industry, the country, the purchasing volume, the difficult specifications, the delivery rhythm. None of that leaks trade secrets, yet it lets a buyer judge whether your situation resembles theirs.
I treat this kind of check as a translation exercise: translating adjectives into facts. If “reliable quality” cannot be translated into a fact, it means you yourself have not thought through where the reliability lies. To see which phrases are an industry-wide habit, compare the risk points in service-provider promises.
Why does empty talk backfire?
The first layer is trust. A buyer looks at ten supplier sites a day, and every one says it leads the field. When every page uses the same set of phrases, those phrases lose their power to differentiate, and the reader’s default reaction is to skip. The harder you push the praise, the more it looks like you are covering something up.
The second layer is how AI retrieves. According to Ahrefs’ 2026-10-12 query fan-out study, one sentence from a user is split into multiple sub-queries, and the model has to find sentences that can support an answer. Adjectives cannot support any answer; specific conditions can. If it is not on your page, you are not in the answer.
The third layer is being held to your own claim. If you write “short lead times” and the customer waits two months after ordering, that page becomes a liability. The risk of empty talk is not only that it does not work; it can also be held up against reality when you are not ready. Writing concrete numbers is pressure, but that pressure is precisely a sign that you are doing things seriously. Combine this with service-provider risk signals to see which promises cannot hold.
One comparison from our own panel
The T0 baseline round was on 2026-09-17 and T1 on 2026-10-12, with 72 answers in each round. Of the 12 answers to provider-type questions, T0 put Tianwen on the list in none of them; T1 did in 2. Strict brand mentions went from 3/72 to 6/72.
What I want to stress is where those 2 appeared. They appeared under questions like “how to choose a service provider” and “why do quotes differ so much,” not on pages where we praised ourselves. The reason the model chose us is that those answers needed a candidate for comparison, not because we wrote “we are professional.”
There is counter-evidence too. In the same round, the total number of cited domains fell from 322 to 245, with an overlap of only 90. If self-promotion could reliably win citations, the source set would not swing that much. The better explanation is that the model keeps looking for usable sources of information, and whoever supplies facts has a chance. I recorded this thread in what characterizes content that AI mentions.
How do you turn self-promotion into something verifiable?
Replace adjectives with ranges and conditions
“Low minimum order quantity” becomes “500 pieces for standard styles, 2,000 for custom colors.” “Fast lead times” becomes “7 days from stock, 25 to 30 days for custom, plus one week in peak season.” With ranges, a buyer can judge whether it suits them, and AI has a sentence to extract. For the structure of the writing, see the four-piece set: definition, table, Q&A, evidence.
Replace results with process
“We are strict about quality control” becomes “the sampling ratio for each batch, how we handle an anomaly when we find one, what caused the anomaly in the last batch.” Process is more credible than results because it is harder to fake. And writing about process forces you to face the real state of your production line, which is a management benefit in itself.
Leave a place where someone can disprove you
Write down both what it suits and what it does not suit. For example, “this process suits small batches with many varieties, not very large single-color runs.” Voluntarily saying what it does not suit actually raises credibility, because the buyer knows you are making the judgment on their behalf.
If you are choosing a service provider, you can use the same standard to check them in reverse. Compare the red flags for AI search service providers; which phrases are empty talk becomes obvious at a glance.
A low-cost action checklist
Three steps, no disruption to normal business, and the first round can be finished within a week.
- Circle every adjective on the home page and the about page: ask of each one, “can this become a number, a condition, or a process?” Delete the ones that cannot.
- Add one fact to each page that only you know: it can be a sampling ratio, a reason for rework, a peak-season delay rule — any single one counts.
- Add the scenario to your customer cases: no client name needed; the industry, the country, the difficult specifications, and the delivery rhythm are enough.
After the changes, do not rush to check rankings. Read it yourself first: if a buyer who has never heard of you read only these three passages, could they tell whether you suit them? To systematize this move, see how to build B2B export visibility in AI search.
The questions owners ask most often
Can a page mention our own company at all?
Yes, and it should. The issue is not whether to mention it but how. Tying the company name to verifiable facts is an introduction; piling on adjectives is self-promotion.
Why does a phrase like “we are experienced” backfire?
Because that sentence adds zero information; anyone can write it. Neither readers nor AI can use it to judge anything, and it actually lowers trust in the whole page.
Do customer cases count as self-promotion?
Not if they carry specific conditions and results. Spelling out the industry, specifications, lead time, the problem encountered, and how it was handled is evidence; writing only “we have served many well-known clients” is self-promotion.
We have already written a lot of self-promotional content. Should we delete it all?
No. Start with the pages that earn impressions, replace the adjectives with verifiable facts, and keep the existing structure. Deleting pages costs far more than revising them.
This page is machine-translated from the Simplified Chinese version and is pending human review.