Yes, but the job links do has shifted. Backlink building for GEO isn't about pushing a page from position 8 to position 3 any more. It's about whether a language model, at the moment it assembles an answer, can find a page that says something specific about your brand on a domain its retrieval layer trusts. Links get you into that pool. They no longer get you the citation on their own.
We've been running SEO for Shopify brands since long before anyone said "generative engine optimisation" out loud, and the honest read from the last eighteen months is that link acquisition still earns its budget line. It just earns it differently, and the shape of a sensible link programme in 2026 looks nothing like the guest-post spreadsheets people were buying in 2019.
What generative engines actually reach for
Every assistant that answers a shopping question is doing one of two things. Either it's recalling something baked into training weights, or it's running a live retrieval and summarising what it pulled back. For commercial queries — "best cold-pressed coconut oil brands in India", "is X worth it compared to Y" — it's almost always retrieval, because the model knows its training data is stale on anything with a price attached.
Retrieval means an index. Google's AI Overviews sit on top of Google's own index. ChatGPT's search has leaned on Bing's. Perplexity, OpenAI and Anthropic all run their own crawlers, and you can see them in your server logs if you look. The practical consequence: if a page isn't crawled, indexed and ranked reasonably well by a conventional search engine, it is not in the candidate set the model summarises from. Links are still one of the main reasons a page gets crawled often and ranked well.
So the chain hasn't broken. It got longer. Link → crawl priority → index position → retrieval candidate → citation. Miss the first step and the last one doesn't happen.
Mentions move the answer more than links do
Here's the part that surprises founders. When we've watched what an assistant actually cites for a category query, the sources are usually a listicle, a Reddit thread, a marketplace review page, a comparison article on a mid-sized publication. The model reads the sentence around your brand name. It doesn't especially care whether that sentence contained an anchor tag.
An unlinked mention that says "the Bengaluru brand makes a fluoride-free version at ₹349" is worth more to a generative answer than a followed link in a footer directory. The mention gives the model a fact it can repeat. The footer link gives it nothing to say.
Which means the old measurement — DA, number of referring domains, link velocity — is measuring the wrong thing for this purpose. What you want to count is: how many pages on the open web contain your brand name inside a sentence that describes what you sell, at what price, for whom. That number is usually much smaller than the referring-domain count, and it's the one that correlates with getting named.
The links that still do work
Four categories, in rough order of how much effort they're worth:
- Category roundups on publications that rank. If "best running shoes under ₹5000" surfaces a particular publication's list on page one, that list is in the retrieval pool for the same question asked to an assistant. Getting into it is worth more than fifty niche blog links.
- Comparison and alternatives pages — including ones on competitors' or aggregators' sites. Uncomfortable, but models love a comparison table because it's structured and answers the question directly.
- Genuine editorial coverage with a specific claim in it. A product launch story that names the material, the price and the founder is retrievable. A press release that says the brand is "revolutionising" a category is not, because there's no fact to extract.
- Community and forum presence. Reddit, Quora, category-specific forums, Indian WhatsApp-adjacent communities that have public web archives. These get cited constantly. You cannot fake your way in, and the attempts are obvious.
Notice what's absent. Directory submissions. Paid link marketplaces. The ₹3,000-per-placement sites that exist purely to sell links and have no readers. Those still get indexed, occasionally, and they still pass a trickle of signal, but no assistant is summarising a post on a site with no traffic and eleven unrelated categories.
Doing the arithmetic before you commit the budget
A concrete case. Say you're weighing ₹1,60,000 a quarter on link building. Option A is forty placements at ₹4,000 each on the usual Indian guest-post inventory. Option B is four earned placements at ₹40,000 each in real publications, including the outreach time.
Run the break-even on referral clicks alone. AOV ₹2,400, contribution margin after COGS, shipping and payment fees of 38% — that's ₹912 per order. To recover ₹1,60,000 you need 175 orders. At a 3% site-wide conversion rate that's roughly 5,850 sessions, or about 490 sessions a month sustained over a year.
Will four editorial placements send you 490 sessions a month? Occasionally one will and the other three won't. Forty guest posts on dead blogs certainly won't; we've seen batches like that produce under 200 sessions in total across six months.
So if you justify link spend purely on referral traffic, both options usually fail the test. The case has to be made on the second-order effect: better crawl frequency, better rankings on the money terms, and your brand name sitting inside sentences that assistants read. Option B does that. Option A does not do it at all. Same budget, and one of them is a write-off.
Run that calculation with your own AOV and margin before anyone sells you a package. Half the link proposals we see fall apart on the first line of arithmetic.
What Shopify stores should be pointing links at
Most brands aim every link at the homepage, which is close to useless for retrieval. The homepage says the brand exists. It rarely answers a question.
Point links at pages that contain extractable facts:
- Collection pages with real editorial copy above or below the grid, not 40 words of keyword mush. A collection page that explains how to choose between the six variants is a page a model can quote.
- Comparison pages you own. "Ours vs the imported alternative" with an actual table of specs and prices. These get cited a lot, partly because so few Indian D2C brands publish them.
- Ingredient, material and care pages. For food and personal care, an FSSAI-compliant ingredient breakdown with sourcing detail is exactly the kind of page that gets pulled into an answer about safety or suitability.
- Size, fit and sizing-conversion guides. Boring, evergreen, endlessly retrieved.
- Your shipping, COD and returns policy page, if it's written in plain sentences. Assistants get asked "do they do cash on delivery" constantly.
One structural note. If your product pages are rendered so heavily by apps that the meaningful text arrives after JavaScript execution, some crawlers will get a thin page. We've had stores where the specification table only existed inside a third-party tab widget, and the assistant answering questions about that product invented the specs. Fixing the render was worth more than the links.
Measuring GEO without kidding yourself
There's no rank tracker for this that we'd trust with a budget decision. What we do instead:
- Server logs. Filter for GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended. Which URLs are they hitting, how often, and are they getting 200s? If a crawler has never touched your comparison page, nothing downstream matters.
- Referral segments in GA4. Small numbers, but the conversion rate on them is usually well above site average, because the person arrived with intent already formed. Track it as a separate channel from day one so you have a baseline.
- Manual prompt panels. Twenty questions a real buyer would ask, run monthly across three assistants, logged in a sheet with the sources each one cited. Tedious. Also the only measurement that tells you what changed and why.
- The post-purchase survey. Add "ChatGPT or another AI assistant" to your how-did-you-hear-about-us options. Self-reported attribution is noisy, and it's still better than nothing.
What we don't do is buy a GEO score from a dashboard. Most of those tools are running the same prompt panel we'd run by hand, with a monthly fee attached.
The uncomfortable bit about "GEO packages"
A lot of what's being sold as generative engine optimisation right now is the 2013 link package with a new cover sheet. Same vendors, same inventory, same reporting spreadsheet with "AI visibility" pasted over the old column header. If a proposal quotes you a fixed number of links per month at a fixed price per link, it isn't a GEO programme. It's link volume, and link volume stopped being the constraint years ago.
The other overrated tactic is stuffing schema everywhere in the hope that structured data forces a citation. Product, FAQ and Organization schema are worth having and take an afternoon on most Shopify themes. They don't override the retrieval layer's judgement about whether your page is a good source. We've watched perfectly marked-up pages get passed over for an unstructured Reddit comment.
Where we'd put ₹1,50,000 a quarter
For a mid-size Indian D2C brand, roughly:
- ₹60,000 on earned placements in three or four publications that actually rank for your category terms — pitching, samples, follow-up, the unglamorous part
- ₹45,000 on producing the comparison, ingredient and buying-guide pages worth linking to, because outreach without a destination page converts badly
- ₹25,000 on fixing the technical floor: render-blocking scripts, thin collection copy, broken canonicals, whatever the log files say is being ignored
- ₹20,000 on genuine community participation, which mostly means someone's time rather than a media spend
Nothing on directories. Nothing on link marketplaces. If the quarter is tight, cut the placements before you cut the page production, because a good comparison page earns links on its own and a link to a weak page earns nothing.
Our SEO and content work for Shopify brands runs on roughly that split, and the balance moves depending on how much substance the site already has. Brands with a deep catalogue and no editorial pages usually need to build first and pitch second.
Start with the logs
Before you spend anything on links, pull thirty days of server logs and check whether the AI crawlers are reaching your key pages at all. It takes an hour and it changes the plan more often than not. If you'd like a second pair of eyes on what they're finding, our free store audit covers crawl behaviour alongside the usual speed and conversion checks, and you can see the sort of thing we look for in our case studies.