From invisible to page one: a two-week diary of one Upwork profile

In two weeks one paid advertising profile went from being findable for 9 of 37 tracked client queries to 29, with its priority query moving from position 49 to position 2. This is the diary of how that happened, in the order it happened, including the three experiments that produced nothing and the mistake that cost us a week of measurements.

The freelancer is a performance marketer in one of the most crowded categories on the platform, and the profile is kept anonymous. The field-by-field findings from this cycle live in a separate reference article — this one is about the order things happened in, and the decisions that were not ours to make.

Days 1–2: refusing to change anything

The first two days went entirely into measurement, which the client found frustrating and which turned out to be the reason the rest worked. We built a list of 37 queries a client in this niche would actually type, then recorded where the profile stood for each one.

The starting picture: found for 9 queries out of 37. Strong on a couple of broad words, absent from most of the exact phrases that bring work. No invitations, and the owner's theory was that the overview was too short.

Days 3–6: the title, and the discovery that it splits rather than lifts

The title is the strongest field, so it went first. We wrote two variants differing by two words, put each live, waited the few minutes it takes for search to update, and measured.

This was the most useful surprise of the whole cycle. We had expected to find a winner. Instead we found a fork, and the choice belonged to the freelancer: which of those two client populations did he want to be in front of? He picked the one matching the contracts he actually enjoys.

We also threw words out of the title. Multi-Country, Growth Partner, an ampersand — nobody searches for those. Freeing those characters bought two more real phrases.

Day 7: the experiment that came back empty

The client was convinced the overview was the problem. We tested it properly: loaded it with target phrasing, measured, stripped it back, measured again. No movement in either direction. We then did the same with the employment history, five separate times, the last one with nine thousand characters of it.

Nothing moved. And the tempting conclusion — *these fields do not work* — is wrong, so we will not draw it. Both fields are indexed: on a separate profile, a rare phrase placed only in the overview brought that profile onto page one for it. What this niche shows is different. With tens of thousands of profiles and the leaders separated by fractions of a point, a quiet signal cannot surface. The field works; the crowd is louder.

What it bought us was permission to stop. The two fields the client most wanted to polish were real, just not where his wins were — and a week of rewriting them would have produced a nicer profile and the same position.

Days 8–11: portfolio, where the volume actually lives

Portfolio item descriptions carry real weight, and we proved it in the least pleasant way: during a cleanup pass we removed keyword occurrences from several items, and the tracked queries lost 199 positions in aggregate. We put them back.

Portfolio titles are indexed too, which we confirmed by planting a nonsense word in one and finding it in search. So each item got a title that carries real phrases and a description written to the field's full length, with the phrases that matter placed first — when the field truncates, you want the tail to fall off, not the keywords.

The honest scale point, which we tell every client before they get excited: nine new portfolio items moved the Google-related queries by a handful of positions. Moving a large direction decisively takes two to three hundred items. That is months of daily work, and saying so upfront is better than implying a shortcut exists.

Days 12–14: the two quiet fields

Skills came next, and they do not work the way most people assume. They barely influence ranking. What they do is decide whether you survive a client's filter: tick a skill the profile does not list, and the profile disappears from the results entirely, even if it was first without the filter.

Then the rate. It has no effect on position whatsoever, and a large effect on who ever sees the profile, because the client's price slider moves in fixed steps of ten. A rate sitting exactly on a step qualifies for the windows on both sides of it; a rate between steps only ever qualifies for the one above. This profile sat between steps, which quietly removed it from the range where most of its market browses. Moving it onto a round number cost nothing and changed who it reached.

What we would do differently

1. 01Three measurement passes from day one, not from day five. Everything before that had to be redone.
2. 02Ship one change per day. Three edits went live together on one day and produced a combined shift we could never decompose.
3. 03Ask about the freelancer's preferred clients before writing title variants, not after discovering the fork.
4. 04Set expectations about portfolio scale in the first conversation — the gap between nine items and three hundred is the gap between a nudge and a move.

The underlying method, written out step by step, is here. The field-by-field evidence behind these numbers is here.