Ask buyers who own a site-search tool what they'd tell their past self and it isn't about the search bar — it's about the work behind it. Whether they budgeted for relevance tuning, whether they had the engineers the powerful tools assume, what the usage meter really cost, and whether the "AI search" they paid for actually beat well-configured conventional search. The platforms differ; the regrets converge on the effort relevance demands. Here they are, in the order buyers hit them — and if you're already looking to leave your current tool, the triggers are in why ecommerce teams switch site-search vendors. Note this is on-site search, distinct from optimizing for AI answer engines, which is its own read: the scramble to be the AI's answer.
The seven lessons
Relevance is a program, not a feature — the tool won't tune itself
The regret underneath all the others. Buyers who expected a search tool to deliver great results out of the box were consistently disappointed: the powerful platforms hand you the controls — synonyms, ranking rules, merchandising overrides, personalization — but not the answers. Relevance has to be configured and maintained against your specific catalog and how your customers actually search, and buyers who didn't staff that work describe capable tools producing mediocre results. Before you shop for a new platform because search is bad, ask whether anyone owns relevance as an ongoing job. Usually the fix is a process, not another vendor.
Named in this context: Algolia · Coveo · Bloomreach
The most-rated pure-play is developer-dependent
The most-rated pure-play in the corpus is beloved for its API, headless flexibility, and integration depth — and, in the same breath, flagged as complex and requiring significant technical expertise. Buyers describe developer confusion over API communication and the order of operations, and difficulty implementing the full range of features without engineering. That's fine if you have the bench; it's a trap if you don't. Match the tool to your team: an API-first platform rewards engineering-rich teams and punishes lean ones. If merchandisers, not developers, will run search day to day, weigh a more turnkey, merchandiser-friendly tool instead.
Named in this context: Algolia · Elastic
Pricing scales against you — and stings smaller teams hardest
Cost is a recurring complaint, and it compounds. Buyers describe the leading platform's pricing as high and rising year over year, enough to be a significant burden, and smaller companies in particular call it "way too expensive for the value it adds." Because usage-based search pricing grows with traffic and queries, the sticker that fit at launch doesn't fit at scale — and the value has to keep pace with the meter. Model the cost against your query volume and growth, not today's traffic, and be honest about whether a smaller catalog justifies a premium platform or would be served by a lighter one.
Named in this context: Algolia
AI search is overpromised — validate the claim on your own catalog
The category is repricing on AI, and buyers are skeptical for good reason. Several report that a leading platform's AI capabilities hadn't advanced enough to meet their expectations, and that features like dynamic reranking didn't work as seamlessly as pitched. The higher-rated enterprise platforms suggest AI does add value when it's mature — but the gap between the AI-search pitch and delivery is wide. Buyers here do run proofs of concept, but they are general procurement exercises rather than search ones — nobody describes benchmarking a vendor's AI against their own catalog and query log before buying, so the skepticism above is a verdict formed after the fact. So the next part is our recommendation rather than theirs: treat "AI relevance" as a claim to test, not a given — during the trial, run the vendor's AI against your real catalog and real queries, and compare it head-to-head with well-tuned conventional search before paying the AI premium.
Named in this context: Algolia · Coveo · Bloomreach
Merchandising control matters as much as raw relevance
Search isn't only an algorithm — it's a merchandising surface. Merchandisers need to pin, boost, bury, and curate results for campaigns and margin without filing an engineering ticket, and the tools differ sharply in how much of that they hand to a non-technical user. Buyers who bought a specialist search platform sometimes found the merchandising controls thinner than they needed, forcing engineering into the loop for routine tuning. Map who will actually operate search and what they need to control day to day, and weight merchandiser self-service as heavily as the underlying relevance engine.
Named in this context: Algolia · Constructor · Searchspring
Match the tool to your team and catalog — enterprise vs. pure-play vs. DIY
There's no single best site search; there's a best fit, and the ratings map to team and catalog rather than to a winner. The enterprise AI-search platforms rate highest for large, complex catalogs with the budget and the need for out-of-box AI relevance. The search pure-play suits engineering-rich teams that want flexibility and control. The DIY, open-source route is powerful and cost-efficient — but it's a toolkit you have to build and run, worth it only if search is a core competency you want to own. The mistake is buying by reputation; the popular default is mid-pack precisely because it's the wrong fit for many of the teams that bought it.
Named in this context: Coveo (enterprise) · Algolia (pure-play) · Elastic (DIY)
Site search overlaps personalization, recommendations, and the platform — map it before you buy
Search rarely sits alone. The leading search tools also push into product recommendations and personalization, and your commerce platform and merchandising suite may already cover part of the job — so a standalone search purchase can overlap tools you already pay for. Buyers who didn't map the overlap ended up with redundant capabilities and integration seams; those who did bought search to fill a specific gap and let one system own recommendations. Before you sign, inventory what your platform, personalization engine, and recommendations tool already do for discovery, and buy the gap rather than a second engine for the same job.
Named in this context: Algolia · Nosto · Bloomreach
How buyers talk about the tools
The landscape splits by who runs search. Algolia — around 6.9 and by far the most-rated — is the search pure-play: praised for its relevance and flexibility, and support relationship, flagged for complexity, cost that rises year over year, and AI that buyers say lags the pitch. Coveo (around 7.5, on a small sample) is the higher-rated enterprise AI-search platform for large, complex catalogs, and Elastic (rated by too few buyers to publish a number) the powerful DIY toolkit for teams that want to own search. Nosto and Bloomreach bundle search with personalization and merchandising, appealing to teams that want discovery in one system rather than a specialist. The through-line: buyers rate these tools on the effort relevance demands and the fit to their team, not on raw capability — which is why the popular default sits mid-pack. For the pure-play against the bundled suite head-to-head, see Algolia vs. Bloomreach Discovery.
The stories behind the lessons
The counter-current: the tool is rarely the whole problem
The buyers happy with their site search did the unglamorous work the demo skips. They put someone in charge of relevance and iterated on it; they matched the tool to their team — enterprise AI for a complex catalog, a pure-play for an engineering-rich team, DIY only if search was core; they tested AI claims on their own catalog; and they mapped the overlap with personalization and the platform before buying. The frustrated buyers — great tool, mediocre results; an AI premium for relevance they still tuned by hand; a bill that outgrew a small catalog — tended to buy capability and skip the process. The lessons above are cheap; re-platforming search to fix a relevance problem the last tool didn't cause is not.
How buyers rate the tools
| Tool | Buyer rating | What buyers say |
|---|---|---|
| Coveo | around 7.5 | Too few buyers for a decimal. The enterprise AI-search option for large, complex catalogs — stronger on out-of-box relevance, heavier and pricier. |
| Constructor.io | around 7.5 | Too few buyers for a decimal. Named where merchandiser self-service weighs as heavily as the relevance engine underneath. |
| Bloomreach | 7.3 | Bundles search with personalization and merchandising, so buyers are rating more than the search box. |
| Algolia | 6.9 | By far the most-rated here, and mid-pack. Praised for its API, headless flexibility and support relationship; flagged for complexity, cost that rises year over year, and AI that buyers say lags the pitch. |
| Nosto | 6.7 | Sits across search, recommendations and personalization — the appeal is discovery in one system rather than a specialist. |
| Searchspring | 6.3 | A mid-market specialist, named alongside Algolia and Constructor.io where merchandiser self-service is the deciding factor. |
| Elastic | not rated | Too few buyers rated it to publish a number. The DIY route for teams with the engineering to own search outright. |
Want this read against your own stack?
Get my read →What this means for your evaluation
Four checks before you sign. First, decide who owns relevance as an ongoing job — if no one will, a better engine won't save your search. Second, match the tool to your team and catalog: enterprise AI search for large complex catalogs with budget, a search pure-play for engineering-rich teams, DIY only if search is core. Third, test the AI on your own catalog and real queries, head-to-head with well-tuned conventional search, before paying the AI premium. Fourth, map the overlap with your commerce platform, personalization engine, and recommendations tool, and buy only the gap. For the different but adjacent question of showing up in AI answer engines, see the scramble to be the AI's answer.
Common questions
Is Algolia worth the cost?
It depends on whether you have the engineering to use it and the scale to justify it. Algolia is the most-rated on-site search platform in Alium's corpus, and buyers praise its API, headless flexibility, and support relationship — but it averages 6.8/10, held back by three recurring complaints: it's complex and needs significant technical expertise, its pricing is high and rises year over year, and several buyers say its AI and personalization haven't advanced enough to meet expectations. Smaller companies in particular describe it as 'way too expensive for the value it adds.' For a team with engineers and a large catalog that will invest in tuning relevance, buyers find it powerful; for a lean team expecting great search out of the box, the cost and complexity often outrun the value. Price it against the relevance work you'll actually staff, not the demo.
Why is my site search bad even after buying a tool?
Because relevance is a program, not a feature you switch on. The most common regret in the interviews is expecting a search tool to deliver great results out of the box. In practice, buyers find that relevance requires ongoing tuning — synonyms, ranking rules, merchandising overrides, and personalization all have to be configured and maintained against your specific catalog and how your customers actually search. The powerful platforms give you the controls but not the answers, and buyers who didn't staff the tuning work describe capable tools producing mediocre results. If your search is underperforming after a purchase, the fix is usually a relevance and merchandising process — someone who owns it and iterates — not another platform.
Is AI site search actually better?
Sometimes, but buyers say the category has overpromised on it, so validate the claim on your own catalog before you buy it. Several buyers report that a leading platform's AI capabilities hadn't advanced enough to meet their expectations, and that features like dynamic reranking didn't work as seamlessly as pitched. Enterprise AI-search platforms rate higher in the corpus than the search pure-plays, which suggests the AI does add value when it's mature — but the gap between the AI-search pitch and delivery is wide enough that buyers should treat it as a claim to test, not a given. Nobody here describes running that benchmark before buying, so this last part is our recommendation rather than a practice buyers report: run the vendor's AI relevance against your real product catalog and real queries during the trial, and compare it head-to-head with well-tuned conventional search before paying the AI premium.
Algolia vs. Coveo vs. Elastic — which site search is best?
They serve different teams, and the ratings reflect it. Algolia (about 6.9/10, and by far the most-rated) is the search pure-play — flexible and strong on relevance, but complex, costly at scale, and dependent on engineering. Coveo (around 7.5, on a small sample) is the higher-rated enterprise AI-search option, stronger on out-of-box AI relevance but heavier and pricier, aimed at large, complex catalogs. Elastic (rated by too few buyers to publish a number) is the DIY route — powerful and cost-efficient if you have the engineering to build and run it, but it's a toolkit, not a turnkey product. The pattern buyers describe: the popular default isn't the best-rated, so match the tool to your team and catalog — enterprise AI search for large complex catalogs with budget, the search pure-play for engineering-rich teams, and DIY only if search is a core competency you want to own.
Get this research made for your stack
Choosing or replacing an on-site search tool right now? Do a 15-minute interview about your catalog and team and get this personalized — what peers with your stack chose, where each tool's relevance, cost, and AI claims hold up, and which of these lessons apply to your shortlist.
Get my personalized read — 15-min interviewNo password · your interview is anonymized before it ever informs a page like this one.