Do Media Backlinks Still Work? The Honest SEO and GEO Answer

Start with what Google actually says

Google's link spam policy is public and specific: links acquired in exchange for payment, goods or services should be qualified with rel="sponsored" or rel="nofollow". That covers paid guest posts, sponsored articles and most of what this industry sells. It is not ambiguous and it has not quietly changed.

So the standard pitch (buy this placement, get a dofollow link from a high-authority domain, watch your rankings move) is selling something that either does not comply with Google's stated policy, or does comply and therefore does not do what the pitch says.

It is worth being blunt about this because the alternative is a purchase built on a misunderstanding. If your entire business case for a placement is the link attribute, you are buying a variable that the publisher can change with one CMS setting, retroactively, across every article they have ever run, without telling you.

What press coverage actually does for search

Brand search volume

The most underrated mechanism in SEO. When people see your name in a publication they read, some of them later type your name into Google. Branded search volume is a signal that behaves very differently from links, cannot be bought directly, and correlates with the kind of visibility that survives algorithm updates. A placement that produces branded searches has done something no link audit will show you.

Being findable by the people who write about your category

Journalists, analysts and bloggers research by searching. If your company appears in indexed articles when someone researches your category, you enter the consideration set for coverage you did not pay for. This compounds: the third article about you is easier to earn than the first, because the first two are what the writer found.

Entity recognition

Google builds an understanding of your company as an entity (what it does, who runs it, what it is associated with) from consistent mentions across sources it trusts. Named, dated, unambiguous mentions in indexed publications feed that directly, whether or not they carry a followed link. This is why a company with strong press presence often outranks a better-optimised competitor on category terms.

Referral traffic

Usually smaller than expected, and worth setting expectations on. A single placement in a national title typically sends a modest number of visitors, concentrated in the first 48 hours. If someone sells you a placement on projected traffic, ask them for the analytics from a comparable one.

The part that changed: answer engines

Until recently, being a source and being a search result were the same thing. That is no longer true. A growing share of questions about your category now get answered inside ChatGPT, Perplexity, Gemini, Claude or Google's AI overviews, and those answers are assembled from sources, with the source often named.

Getting named in that answer is a different optimisation problem from ranking, and the industry has a name for it now: Generative Engine Optimization, or GEO. What we observe consistently is that language models reach for sources with a specific profile:

• Named and dated. An article with a byline and a publication date reads as accountable. Undated evergreen pages do not.
• Factually dense. Specific numbers, named entities, concrete mechanisms. Vague marketing prose gets skipped because there is nothing in it to cite.
• Published somewhere with a reputation. The domain still matters, for the same reason it matters to a human researcher.
• Structured to be extracted. Clear headings, direct answers near the question, tables, explicit definitions.
• Indexed and crawlable. If it is not indexed, it does not exist for either kind of engine.

Notice what is absent from that list: the link attribute. As far as we can observe, whether a link carries rel="sponsored" has no bearing on whether a model cites the article containing it. What matters is whether the article says something specific enough to be worth quoting.

How to buy coverage that gets cited

If citation is the goal, some of the standard advice inverts. Six things that actually matter:

1. Insist the article is indexed, and verify it. Not "published", but indexed. Search a distinctive sentence in Google a week later. An article that never gets indexed is invisible to every engine that matters.
2. Put numbers in it. Your own data, your own benchmarks, your own survey. A model cannot cite an adjective. Proprietary numbers are the most citable thing you own.
3. Answer a real question in the headline and the first paragraph. Extraction favours proximity: the answer should sit next to the question, not four scrolls below it.
4. Be consistently named. Same company name, same spelling, same founder attribution everywhere. Inconsistency fragments the entity and dilutes every mention.
5. Choose titles that are actually crawled. Some sites block AI crawlers in robots.txt. A placement in a title that blocks GPTBot cannot be cited by ChatGPT, whatever it cost you.
6. Spread across sources rather than doubling down on one. Corroboration across several independent publications does more for a model's confidence than one prestigious mention.

Five mistakes that waste the budget

1. Buying domain rating instead of readership. A site can score 80 on an SEO tool and have no human audience whatsoever. Those sites are cheap because that is what they are worth. Read three of the outlet's non-sponsored articles before you buy anything.
2. Never checking indexing. A surprising share of purchased placements never get indexed, usually because they sit in a section the publisher deprioritises. Nobody notices, because nobody checks. Search a distinctive sentence a week after publication. It takes fifteen seconds.
3. Writing the article for the algorithm instead of the reader. Keyword-stuffed placement copy is worse on every axis: readers bounce, editors resent it, and models skip it because there is no fact in it worth extracting.
4. Buying one expensive placement instead of four useful ones. Corroboration across independent sources builds an entity. A single prestige mention builds a screenshot.
5. Measuring nothing. If you cannot say what your branded search volume was before and after, or whether an assistant names you now and did not six months ago, you have no way to tell a good supplier from a lucky one.

The test you can run this afternoon

Open ChatGPT, Perplexity and Google's AI overview. Ask each one the question your best customer would ask before they knew you existed: "what are the best tools for X", "who does Y for mid-market companies in Europe". Write down which companies get named and which sources the answers cite.

That list is your real competitive set for the next few years, and the sources cited are your actual target list for placements. It costs nothing, takes twenty minutes, and it is a better brief for a press strategy than most agency audits.

Then run it again in three months. This is the closest thing to a rank tracker that currently exists for answer engines, and the companies that start measuring it now will spend the next two years being surprised considerably less often than everyone else.